Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
import json
|
2024-12-15 14:40:03 +01:00
|
|
|
from datetime import datetime, timezone
|
|
|
|
|
from typing import Any, ClassVar, List, Optional, Union
|
|
|
|
|
|
|
|
|
|
import numpy as np
|
|
|
|
|
import pandas as pd
|
|
|
|
|
import pendulum
|
|
|
|
|
import pytest
|
|
|
|
|
from pydantic import Field, ValidationError
|
|
|
|
|
|
|
|
|
|
from akkudoktoreos.config.configabc import SettingsBaseModel
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
from akkudoktoreos.core.coreabc import get_ems
|
2024-12-15 14:40:03 +01:00
|
|
|
from akkudoktoreos.core.dataabc import (
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
DataABC,
|
2024-12-15 14:40:03 +01:00
|
|
|
DataContainer,
|
|
|
|
|
DataImportProvider,
|
|
|
|
|
DataProvider,
|
|
|
|
|
DataRecord,
|
|
|
|
|
DataSequence,
|
|
|
|
|
)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
from akkudoktoreos.core.databaseabc import DatabaseTimestamp
|
2024-12-15 14:40:03 +01:00
|
|
|
from akkudoktoreos.utils.datetimeutil import compare_datetimes, to_datetime, to_duration
|
|
|
|
|
|
|
|
|
|
# Derived classes for testing
|
|
|
|
|
# ---------------------------
|
|
|
|
|
|
|
|
|
|
class DerivedConfig(SettingsBaseModel):
|
|
|
|
|
env_var: Optional[int] = Field(default=None, description="Test config by environment var")
|
|
|
|
|
instance_field: Optional[str] = Field(default=None, description="Test config by instance field")
|
|
|
|
|
class_constant: Optional[int] = Field(default=None, description="Test config by class constant")
|
|
|
|
|
|
|
|
|
|
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
class DerivedBase(DataABC):
|
2024-12-15 14:40:03 +01:00
|
|
|
instance_field: Optional[str] = Field(default=None, description="Field Value")
|
|
|
|
|
class_constant: ClassVar[int] = 30
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class DerivedRecord(DataRecord):
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
"""Date Record derived from base class DataRecord.
|
|
|
|
|
|
|
|
|
|
The derived data record got the
|
|
|
|
|
- `data_value` field and the
|
|
|
|
|
- `dish_washer_emr`, `solar_power`, `temp` configurable field like data.
|
|
|
|
|
"""
|
|
|
|
|
|
2024-12-15 14:40:03 +01:00
|
|
|
data_value: Optional[float] = Field(default=None, description="Data Value")
|
|
|
|
|
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
@classmethod
|
|
|
|
|
def configured_data_keys(cls) -> Optional[list[str]]:
|
|
|
|
|
return ["dish_washer_emr", "solar_power", "temp"]
|
|
|
|
|
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
class DerivedSequence(DataSequence):
|
|
|
|
|
# overload
|
|
|
|
|
records: List[DerivedRecord] = Field(
|
|
|
|
|
default_factory=list, description="List of DerivedRecord records"
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
def record_class(cls) -> Any:
|
|
|
|
|
return DerivedRecord
|
|
|
|
|
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
class DerivedSequence2(DataSequence):
|
|
|
|
|
# overload
|
|
|
|
|
records: List[DerivedRecord] = Field(
|
|
|
|
|
default_factory=list, description="List of DerivedRecord records"
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
def record_class(cls) -> Any:
|
|
|
|
|
return DerivedRecord
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
class DerivedDataProvider(DataProvider):
|
|
|
|
|
"""A concrete subclass of DataProvider for testing purposes."""
|
|
|
|
|
|
|
|
|
|
# overload
|
|
|
|
|
records: List[DerivedRecord] = Field(
|
|
|
|
|
default_factory=list, description="List of DerivedRecord records"
|
|
|
|
|
)
|
|
|
|
|
provider_enabled: ClassVar[bool] = False
|
|
|
|
|
provider_updated: ClassVar[bool] = False
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
def record_class(cls) -> Any:
|
|
|
|
|
return DerivedRecord
|
|
|
|
|
|
|
|
|
|
# Implement abstract methods for test purposes
|
|
|
|
|
def provider_id(self) -> str:
|
|
|
|
|
return "DerivedDataProvider"
|
|
|
|
|
|
|
|
|
|
def enabled(self) -> bool:
|
|
|
|
|
return self.provider_enabled
|
|
|
|
|
|
|
|
|
|
def _update_data(self, force_update: Optional[bool] = False) -> None:
|
|
|
|
|
# Simulate update logic
|
|
|
|
|
DerivedDataProvider.provider_updated = True
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class DerivedDataImportProvider(DataImportProvider):
|
|
|
|
|
"""A concrete subclass of DataImportProvider for testing purposes."""
|
|
|
|
|
|
|
|
|
|
# overload
|
|
|
|
|
records: List[DerivedRecord] = Field(
|
|
|
|
|
default_factory=list, description="List of DerivedRecord records"
|
|
|
|
|
)
|
|
|
|
|
provider_enabled: ClassVar[bool] = False
|
|
|
|
|
provider_updated: ClassVar[bool] = False
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
def record_class(cls) -> Any:
|
|
|
|
|
return DerivedRecord
|
|
|
|
|
|
|
|
|
|
# Implement abstract methods for test purposes
|
|
|
|
|
def provider_id(self) -> str:
|
|
|
|
|
return "DerivedDataImportProvider"
|
|
|
|
|
|
|
|
|
|
def enabled(self) -> bool:
|
|
|
|
|
return self.provider_enabled
|
|
|
|
|
|
|
|
|
|
def _update_data(self, force_update: Optional[bool] = False) -> None:
|
|
|
|
|
# Simulate update logic
|
|
|
|
|
DerivedDataImportProvider.provider_updated = True
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class DerivedDataContainer(DataContainer):
|
|
|
|
|
providers: List[Union[DerivedDataProvider, DataProvider]] = Field(
|
|
|
|
|
default_factory=list, description="List of data providers"
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# Tests
|
|
|
|
|
# ----------
|
|
|
|
|
|
|
|
|
|
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
class TestDataABC:
|
2024-12-15 14:40:03 +01:00
|
|
|
@pytest.fixture
|
2024-12-30 13:41:39 +01:00
|
|
|
def base(self):
|
2024-12-15 14:40:03 +01:00
|
|
|
# Provide default values for configuration
|
|
|
|
|
derived = DerivedBase()
|
|
|
|
|
return derived
|
|
|
|
|
|
|
|
|
|
def test_get_config_value_key_error(self, base):
|
|
|
|
|
with pytest.raises(AttributeError):
|
|
|
|
|
base.config.non_existent_key
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class TestDataRecord:
|
|
|
|
|
def create_test_record(self, date, value):
|
|
|
|
|
"""Helper function to create a test DataRecord."""
|
|
|
|
|
return DerivedRecord(date_time=date, data_value=value)
|
|
|
|
|
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
@pytest.fixture
|
|
|
|
|
def record(self):
|
|
|
|
|
"""Fixture to create a sample DerivedDataRecord with some data set."""
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
rec = DerivedRecord(date_time=to_datetime("1967-01-11"), data_value=10.0)
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
rec.configured_data = {"dish_washer_emr": 123.0, "solar_power": 456.0}
|
|
|
|
|
return rec
|
|
|
|
|
|
2024-12-15 14:40:03 +01:00
|
|
|
def test_getitem(self):
|
|
|
|
|
record = self.create_test_record(datetime(2024, 1, 3, tzinfo=timezone.utc), 10.0)
|
|
|
|
|
assert record["data_value"] == 10.0
|
|
|
|
|
|
|
|
|
|
def test_setitem(self):
|
|
|
|
|
record = self.create_test_record(datetime(2024, 1, 3, tzinfo=timezone.utc), 10.0)
|
|
|
|
|
record["data_value"] = 20.0
|
|
|
|
|
assert record.data_value == 20.0
|
|
|
|
|
|
|
|
|
|
def test_delitem(self):
|
|
|
|
|
record = self.create_test_record(datetime(2024, 1, 3, tzinfo=timezone.utc), 10.0)
|
|
|
|
|
record.data_value = 20.0
|
|
|
|
|
del record["data_value"]
|
|
|
|
|
assert record.data_value is None
|
|
|
|
|
|
|
|
|
|
def test_len(self):
|
|
|
|
|
record = self.create_test_record(datetime(2024, 1, 3, tzinfo=timezone.utc), 10.0)
|
|
|
|
|
record.date_time = None
|
|
|
|
|
record.data_value = 20.0
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
assert len(record) == 5 # 2 regular fields + 3 configured data "fields"
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_to_dict(self):
|
|
|
|
|
record = self.create_test_record(datetime(2024, 1, 3, tzinfo=timezone.utc), 10.0)
|
|
|
|
|
record.data_value = 20.0
|
|
|
|
|
record_dict = record.to_dict()
|
|
|
|
|
assert "data_value" in record_dict
|
|
|
|
|
assert record_dict["data_value"] == 20.0
|
|
|
|
|
record2 = DerivedRecord.from_dict(record_dict)
|
2025-06-10 22:00:28 +02:00
|
|
|
assert record2.model_dump() == record.model_dump()
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_to_json(self):
|
|
|
|
|
record = self.create_test_record(datetime(2024, 1, 3, tzinfo=timezone.utc), 10.0)
|
|
|
|
|
record.data_value = 20.0
|
|
|
|
|
json_str = record.to_json()
|
|
|
|
|
assert "data_value" in json_str
|
|
|
|
|
assert "20.0" in json_str
|
|
|
|
|
record2 = DerivedRecord.from_json(json_str)
|
2025-06-10 22:00:28 +02:00
|
|
|
assert record2.model_dump() == record.model_dump()
|
2024-12-15 14:40:03 +01:00
|
|
|
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
def test_record_keys_includes_configured_data_keys(self, record):
|
|
|
|
|
"""Ensure record_keys includes all configured configured data keys."""
|
|
|
|
|
assert set(record.record_keys()) >= set(record.configured_data_keys())
|
|
|
|
|
|
|
|
|
|
def test_record_keys_writable_includes_configured_data_keys(self, record):
|
|
|
|
|
"""Ensure record_keys_writable includes all configured configured data keys."""
|
|
|
|
|
assert set(record.record_keys_writable()) >= set(record.configured_data_keys())
|
|
|
|
|
|
|
|
|
|
def test_getitem_existing_field(self, record):
|
|
|
|
|
"""Test that __getitem__ returns correct value for existing native field."""
|
|
|
|
|
record.date_time = "2024-01-01T00:00:00+00:00"
|
|
|
|
|
assert record["date_time"] is not None
|
|
|
|
|
|
|
|
|
|
def test_getitem_existing_configured_data(self, record):
|
|
|
|
|
"""Test that __getitem__ retrieves existing configured data values."""
|
|
|
|
|
assert record["dish_washer_emr"] == 123.0
|
|
|
|
|
assert record["solar_power"] == 456.0
|
|
|
|
|
|
|
|
|
|
def test_getitem_missing_configured_data_returns_none(self, record):
|
|
|
|
|
"""Test that __getitem__ returns None for missing but known configured data keys."""
|
|
|
|
|
assert record["temp"] is None
|
|
|
|
|
|
|
|
|
|
def test_getitem_raises_keyerror(self, record):
|
|
|
|
|
"""Test that __getitem__ raises KeyError for completely unknown keys."""
|
|
|
|
|
with pytest.raises(KeyError):
|
|
|
|
|
_ = record["nonexistent"]
|
|
|
|
|
|
|
|
|
|
def test_setitem_field(self, record):
|
|
|
|
|
"""Test setting a native field using __setitem__."""
|
|
|
|
|
record["date_time"] = "2025-01-01T12:00:00+00:00"
|
|
|
|
|
assert str(record.date_time).startswith("2025-01-01")
|
|
|
|
|
|
|
|
|
|
def test_setitem_configured_data(self, record):
|
|
|
|
|
"""Test setting a known configured data key using __setitem__."""
|
|
|
|
|
record["temp"] = 25.5
|
|
|
|
|
assert record.configured_data["temp"] == 25.5
|
|
|
|
|
|
|
|
|
|
def test_setitem_invalid_key_raises(self, record):
|
|
|
|
|
"""Test that __setitem__ raises KeyError for unknown keys."""
|
|
|
|
|
with pytest.raises(KeyError):
|
|
|
|
|
record["unknown_key"] = 123
|
|
|
|
|
|
|
|
|
|
def test_delitem_field(self, record):
|
|
|
|
|
"""Test deleting a native field using __delitem__."""
|
|
|
|
|
record["date_time"] = "2025-01-01T12:00:00+00:00"
|
|
|
|
|
del record["date_time"]
|
|
|
|
|
assert record.date_time is None
|
|
|
|
|
|
|
|
|
|
def test_delitem_configured_data(self, record):
|
|
|
|
|
"""Test deleting a known configured data key using __delitem__."""
|
|
|
|
|
del record["solar_power"]
|
|
|
|
|
assert "solar_power" not in record.configured_data
|
|
|
|
|
|
|
|
|
|
def test_delitem_unknown_raises(self, record):
|
|
|
|
|
"""Test that __delitem__ raises KeyError for unknown keys."""
|
|
|
|
|
with pytest.raises(KeyError):
|
|
|
|
|
del record["nonexistent"]
|
|
|
|
|
|
|
|
|
|
def test_attribute_get_existing_field(self, record):
|
|
|
|
|
"""Test accessing a native field via attribute."""
|
|
|
|
|
record.date_time = "2025-01-01T12:00:00+00:00"
|
|
|
|
|
assert record.date_time is not None
|
|
|
|
|
|
|
|
|
|
def test_attribute_get_existing_configured_data(self, record):
|
|
|
|
|
"""Test accessing an existing configured data via attribute."""
|
|
|
|
|
assert record.dish_washer_emr == 123.0
|
|
|
|
|
|
|
|
|
|
def test_attribute_get_missing_configured_data(self, record):
|
|
|
|
|
"""Test accessing a missing but known configured data returns None."""
|
|
|
|
|
assert record.temp is None
|
|
|
|
|
|
|
|
|
|
def test_attribute_get_invalid_raises(self, record):
|
|
|
|
|
"""Test accessing an unknown attribute raises AttributeError."""
|
|
|
|
|
with pytest.raises(AttributeError):
|
|
|
|
|
_ = record.nonexistent
|
|
|
|
|
|
|
|
|
|
def test_attribute_set_existing_field(self, record):
|
|
|
|
|
"""Test setting a native field via attribute."""
|
|
|
|
|
record.date_time = "2025-06-25T12:00:00+00:00"
|
|
|
|
|
assert record.date_time is not None
|
|
|
|
|
|
|
|
|
|
def test_attribute_set_existing_configured_data(self, record):
|
|
|
|
|
"""Test setting a known configured data key via attribute."""
|
|
|
|
|
record.temp = 99.9
|
|
|
|
|
assert record.configured_data["temp"] == 99.9
|
|
|
|
|
|
|
|
|
|
def test_attribute_set_invalid_raises(self, record):
|
|
|
|
|
"""Test setting an unknown attribute raises AttributeError."""
|
|
|
|
|
with pytest.raises(AttributeError):
|
|
|
|
|
record.invalid = 123
|
|
|
|
|
|
|
|
|
|
def test_delattr_field(self, record):
|
|
|
|
|
"""Test deleting a native field via attribute."""
|
|
|
|
|
record.date_time = "2025-06-25T12:00:00+00:00"
|
|
|
|
|
del record.date_time
|
|
|
|
|
assert record.date_time is None
|
|
|
|
|
|
|
|
|
|
def test_delattr_configured_data(self, record):
|
|
|
|
|
"""Test deleting a known configured data key via attribute."""
|
|
|
|
|
record.temp = 88.0
|
|
|
|
|
del record.temp
|
|
|
|
|
assert "temp" not in record.configured_data
|
|
|
|
|
|
|
|
|
|
def test_delattr_ignored_missing_configured_data_key(self, record):
|
|
|
|
|
"""Test deleting a known configured data key that was never set is a no-op."""
|
|
|
|
|
del record.temp
|
|
|
|
|
assert "temp" not in record.configured_data
|
|
|
|
|
|
|
|
|
|
def test_len_and_iter(self, record):
|
|
|
|
|
"""Test that __len__ and __iter__ behave as expected."""
|
|
|
|
|
keys = list(iter(record))
|
|
|
|
|
assert set(record.record_keys_writable()) == set(keys)
|
|
|
|
|
assert len(record) == len(keys)
|
|
|
|
|
|
|
|
|
|
def test_in_operator_includes_configured_data(self, record):
|
|
|
|
|
"""Test that 'in' operator includes configured data keys."""
|
|
|
|
|
assert "dish_washer_emr" in record
|
|
|
|
|
assert "temp" in record # known key, even if not yet set
|
|
|
|
|
assert "nonexistent" not in record
|
|
|
|
|
|
|
|
|
|
def test_hasattr_behavior(self, record):
|
|
|
|
|
"""Test that hasattr returns True for fields and known configured dataWs."""
|
|
|
|
|
assert hasattr(record, "date_time")
|
|
|
|
|
assert hasattr(record, "dish_washer_emr")
|
|
|
|
|
assert hasattr(record, "temp") # allowed, even if not yet set
|
|
|
|
|
assert not hasattr(record, "nonexistent")
|
|
|
|
|
|
|
|
|
|
def test_model_validate_roundtrip(self, record):
|
|
|
|
|
"""Test that MeasurementDataRecord can be serialized and revalidated."""
|
|
|
|
|
dumped = record.model_dump()
|
|
|
|
|
restored = DerivedRecord.model_validate(dumped)
|
|
|
|
|
assert restored.dish_washer_emr == 123.0
|
|
|
|
|
assert restored.solar_power == 456.0
|
|
|
|
|
assert restored.temp is None # not set
|
|
|
|
|
|
|
|
|
|
def test_copy_preserves_configured_data(self, record):
|
|
|
|
|
"""Test that copying preserves configured data values."""
|
|
|
|
|
record.temp = 22.2
|
|
|
|
|
copied = record.model_copy()
|
|
|
|
|
assert copied.dish_washer_emr == 123.0
|
|
|
|
|
assert copied.temp == 22.2
|
|
|
|
|
assert copied is not record
|
|
|
|
|
|
|
|
|
|
def test_equality_includes_configured_data(self, record):
|
|
|
|
|
"""Test that equality includes the `configured data` content."""
|
|
|
|
|
other = record.model_copy()
|
|
|
|
|
assert record == other
|
|
|
|
|
|
|
|
|
|
def test_inequality_differs_with_configured_data(self, record):
|
|
|
|
|
"""Test that records with different configured datas are not equal."""
|
|
|
|
|
other = record.model_copy(deep=True)
|
|
|
|
|
# Modify one configured data value in the copy
|
|
|
|
|
other.configured_data["dish_washer_emr"] = 999.9
|
|
|
|
|
assert record != other
|
|
|
|
|
|
|
|
|
|
def test_in_operator_for_configured_data_and_fields(self, record):
|
|
|
|
|
"""Ensure 'in' works for both fields and configured configured data keys."""
|
|
|
|
|
assert "dish_washer_emr" in record
|
|
|
|
|
assert "solar_power" in record
|
|
|
|
|
assert "date_time" in record # standard field
|
|
|
|
|
assert "temp" in record # allowed but not yet set
|
|
|
|
|
assert "unknown" not in record
|
|
|
|
|
|
|
|
|
|
def test_hasattr_equivalence_to_getattr(self, record):
|
|
|
|
|
"""hasattr should return True for all valid keys/configured datas."""
|
|
|
|
|
assert hasattr(record, "dish_washer_emr")
|
|
|
|
|
assert hasattr(record, "temp")
|
|
|
|
|
assert hasattr(record, "date_time")
|
|
|
|
|
assert not hasattr(record, "nonexistent")
|
|
|
|
|
|
|
|
|
|
def test_dir_includes_configured_data_keys(self, record):
|
|
|
|
|
"""`dir(record)` should include configured data keys for introspection.
|
|
|
|
|
It shall not include the internal 'configured datas' attribute.
|
|
|
|
|
"""
|
|
|
|
|
keys = dir(record)
|
|
|
|
|
assert "configured datas" not in keys
|
|
|
|
|
for key in record.configured_data_keys():
|
|
|
|
|
assert key in keys
|
|
|
|
|
|
|
|
|
|
def test_init_configured_field_like_data_applies_before_model_init(self):
|
|
|
|
|
"""Test that keys listed in `_configured_data_keys` are moved to `configured_data` at init time."""
|
|
|
|
|
record = DerivedRecord(
|
|
|
|
|
date_time="2024-01-03T00:00:00+00:00",
|
|
|
|
|
data_value=42.0,
|
|
|
|
|
dish_washer_emr=111.1,
|
|
|
|
|
solar_power=222.2,
|
|
|
|
|
temp=333.3 # assume `temp` is also a valid configured key
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert record.data_value == 42.0
|
|
|
|
|
assert record.configured_data == {
|
|
|
|
|
"dish_washer_emr": 111.1,
|
|
|
|
|
"solar_power": 222.2,
|
|
|
|
|
"temp": 333.3,
|
|
|
|
|
}
|
|
|
|
|
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
class TestDataSequence:
|
|
|
|
|
@pytest.fixture
|
|
|
|
|
def sequence(self):
|
|
|
|
|
sequence0 = DerivedSequence()
|
|
|
|
|
assert len(sequence0) == 0
|
|
|
|
|
return sequence0
|
|
|
|
|
|
|
|
|
|
@pytest.fixture
|
|
|
|
|
def sequence2(self):
|
|
|
|
|
sequence = DerivedSequence()
|
|
|
|
|
record1 = self.create_test_record(datetime(1970, 1, 1), 1970)
|
|
|
|
|
record2 = self.create_test_record(datetime(1971, 1, 1), 1971)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
2024-12-15 14:40:03 +01:00
|
|
|
assert len(sequence) == 2
|
|
|
|
|
return sequence
|
|
|
|
|
|
|
|
|
|
def create_test_record(self, date, value):
|
|
|
|
|
"""Helper function to create a test DataRecord."""
|
|
|
|
|
return DerivedRecord(date_time=date, data_value=value)
|
|
|
|
|
|
|
|
|
|
# Test cases
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
@pytest.mark.parametrize("tz_name", ["UTC", "Europe/Berlin", "Atlantic/Canary"])
|
|
|
|
|
def test_min_max_datetime_timezone_and_order(self, sequence, tz_name, monkeypatch, config_eos):
|
|
|
|
|
# Monkeypatch the read-only timezone property
|
|
|
|
|
monkeypatch.setattr(config_eos.general.__class__, "timezone", property(lambda self: tz_name))
|
|
|
|
|
|
|
|
|
|
# Create timezone-aware datetimes using the patched config
|
|
|
|
|
dt_early = to_datetime("2024-01-01T00:00:00", in_timezone=config_eos.general.timezone)
|
|
|
|
|
dt_late = to_datetime("2024-01-02T00:00:00", in_timezone=config_eos.general.timezone)
|
|
|
|
|
|
|
|
|
|
# Insert in reverse order to verify sorting
|
|
|
|
|
record1 = self.create_test_record(dt_late, 1)
|
|
|
|
|
record2 = self.create_test_record(dt_early, 2)
|
|
|
|
|
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
|
|
|
|
|
min_dt = sequence.min_datetime
|
|
|
|
|
max_dt = sequence.max_datetime
|
|
|
|
|
|
|
|
|
|
# --- Basic correctness ---
|
|
|
|
|
assert min_dt == dt_early
|
|
|
|
|
assert max_dt == dt_late
|
|
|
|
|
|
|
|
|
|
# --- Must be timezone aware ---
|
|
|
|
|
assert min_dt.tzinfo is not None
|
|
|
|
|
assert max_dt.tzinfo is not None
|
|
|
|
|
|
|
|
|
|
# --- Must preserve timezone ---
|
|
|
|
|
assert min_dt.tzinfo.name == tz_name
|
|
|
|
|
assert max_dt.tzinfo.name == tz_name
|
|
|
|
|
|
|
|
|
|
|
2024-12-15 14:40:03 +01:00
|
|
|
def test_getitem(self, sequence):
|
|
|
|
|
assert len(sequence) == 0
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
dt = to_datetime("2024-01-01 00:00:00")
|
|
|
|
|
record = self.create_test_record(dt, 0)
|
2024-12-15 14:40:03 +01:00
|
|
|
sequence.insert_by_datetime(record)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
assert isinstance(sequence.get_by_datetime(dt), DerivedRecord)
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_setitem(self, sequence2):
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
dt = to_datetime("2024-01-03", in_timezone="UTC")
|
|
|
|
|
record = self.create_test_record(dt, 1)
|
|
|
|
|
sequence2.insert_by_datetime(record)
|
|
|
|
|
assert sequence2.records[2].date_time == dt
|
|
|
|
|
|
|
|
|
|
def test_insert_reversed_date_record(self, sequence2):
|
|
|
|
|
dt1 = to_datetime("2023-11-05", in_timezone="UTC")
|
|
|
|
|
dt2 = to_datetime("2024-01-03", in_timezone="UTC")
|
|
|
|
|
record1 = self.create_test_record(dt2, 0.8)
|
|
|
|
|
record2 = self.create_test_record(dt1, 0.9) # reversed date
|
|
|
|
|
sequence2.insert_by_datetime(record1)
|
|
|
|
|
assert sequence2.records[2].date_time == dt2
|
|
|
|
|
sequence2.insert_by_datetime(record2)
|
|
|
|
|
assert len(sequence2) == 4
|
|
|
|
|
assert sequence2.records[2] == record2
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_insert_duplicate_date_record(self, sequence):
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
dt1 = to_datetime("2023-11-05")
|
|
|
|
|
record1 = self.create_test_record(dt1, 0.8)
|
|
|
|
|
record2 = self.create_test_record(dt1, 0.9) # Duplicate date
|
2024-12-15 14:40:03 +01:00
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
assert len(sequence) == 1
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
assert sequence.get_by_datetime(dt1).data_value == 0.9 # Record should have merged with new value
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_key_to_series(self, sequence):
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
dt = to_datetime(datetime(2023, 11, 6))
|
|
|
|
|
record = self.create_test_record(dt, 0.8)
|
|
|
|
|
sequence.insert_by_datetime(record)
|
2024-12-15 14:40:03 +01:00
|
|
|
series = sequence.key_to_series("data_value")
|
|
|
|
|
assert isinstance(series, pd.Series)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
|
|
|
|
|
retrieved_record = sequence.get_by_datetime(dt)
|
|
|
|
|
assert retrieved_record is not None
|
|
|
|
|
assert retrieved_record.data_value == 0.8
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_key_from_series(self, sequence):
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
dt1 = to_datetime(datetime(2023, 11, 5))
|
|
|
|
|
dt2 = to_datetime(datetime(2023, 11, 6))
|
|
|
|
|
|
2024-12-15 14:40:03 +01:00
|
|
|
series = pd.Series(
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
data=[0.8, 0.9], index=pd.to_datetime([dt1, dt2])
|
2024-12-15 14:40:03 +01:00
|
|
|
)
|
|
|
|
|
sequence.key_from_series("data_value", series)
|
|
|
|
|
assert len(sequence) == 2
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
|
|
|
|
|
record1 = sequence.get_by_datetime(dt1)
|
|
|
|
|
assert record1 is not None
|
|
|
|
|
assert record1.data_value == 0.8
|
|
|
|
|
|
|
|
|
|
record2 = sequence.get_by_datetime(dt2)
|
|
|
|
|
assert record2 is not None
|
|
|
|
|
assert record2.data_value == 0.9
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_key_to_array(self, sequence):
|
|
|
|
|
interval = to_duration("1 day")
|
|
|
|
|
start_datetime = to_datetime("2023-11-6")
|
|
|
|
|
last_datetime = to_datetime("2023-11-8")
|
|
|
|
|
end_datetime = to_datetime("2023-11-9")
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
|
|
|
|
|
record1 = self.create_test_record(start_datetime, float(start_datetime.day))
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
record2 = self.create_test_record(last_datetime, float(last_datetime.day))
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
|
|
|
|
|
retrieved_record1 = sequence.get_by_datetime(start_datetime)
|
|
|
|
|
assert retrieved_record1 is not None
|
|
|
|
|
assert retrieved_record1.data_value == 6.0
|
|
|
|
|
|
|
|
|
|
retrieved_record2 = sequence.get_by_datetime(last_datetime)
|
|
|
|
|
assert retrieved_record2 is not None
|
|
|
|
|
assert retrieved_record2.data_value == 8.0
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
series = sequence.key_to_series(
|
|
|
|
|
key="data_value", start_datetime=start_datetime, end_datetime=end_datetime
|
|
|
|
|
)
|
|
|
|
|
assert len(series) == 2
|
|
|
|
|
assert series[to_datetime("2023-11-6")] == 6
|
|
|
|
|
assert series[to_datetime("2023-11-8")] == 8
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=start_datetime,
|
|
|
|
|
end_datetime=end_datetime,
|
|
|
|
|
interval=interval,
|
|
|
|
|
)
|
|
|
|
|
assert isinstance(array, np.ndarray)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
np.testing.assert_equal(array, [6.0, 7.0, 8.0])
|
2024-12-15 14:40:03 +01:00
|
|
|
|
2024-12-29 18:42:49 +01:00
|
|
|
def test_key_to_array_linear_interpolation(self, sequence):
|
|
|
|
|
"""Test key_to_array with linear interpolation for numeric data."""
|
|
|
|
|
interval = to_duration("1 hour")
|
|
|
|
|
record1 = self.create_test_record(pendulum.datetime(2023, 11, 6, 0), 0.8)
|
|
|
|
|
record2 = self.create_test_record(pendulum.datetime(2023, 11, 6, 2), 1.0) # Gap of 2 hours
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=pendulum.datetime(2023, 11, 6),
|
|
|
|
|
end_datetime=pendulum.datetime(2023, 11, 6, 3),
|
|
|
|
|
interval=interval,
|
|
|
|
|
fill_method="linear",
|
|
|
|
|
)
|
|
|
|
|
assert len(array) == 3
|
|
|
|
|
assert array[0] == 0.8
|
|
|
|
|
assert array[1] == 0.9 # Interpolated value
|
|
|
|
|
assert array[2] == 1.0
|
|
|
|
|
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
|
|
|
|
|
def test_key_to_array_linear_interpolation_out_of_grid(self, sequence):
|
|
|
|
|
"""Test key_to_array with linear interpolation out of grid."""
|
|
|
|
|
interval = to_duration("1 hour")
|
|
|
|
|
start_datetime= to_datetime("2023-11-06T00:30:00") # out of grid
|
|
|
|
|
end_datetime=to_datetime("2023-11-06T01:30:00") # out of grid
|
|
|
|
|
|
|
|
|
|
record1_datetime = to_datetime("2023-11-06T00:00:00")
|
|
|
|
|
record1 = self.create_test_record(record1_datetime, 1.0)
|
|
|
|
|
|
|
|
|
|
record2_datetime = to_datetime("2023-11-06T02:00:00")
|
|
|
|
|
record2 = self.create_test_record(record2_datetime, 2.0) # Gap of 2 hours
|
|
|
|
|
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
|
|
|
|
|
# Check test setup
|
|
|
|
|
record1_timestamp = DatabaseTimestamp.from_datetime(record1_datetime)
|
|
|
|
|
record2_timestamp = DatabaseTimestamp.from_datetime(record2_datetime)
|
|
|
|
|
start_timestamp = DatabaseTimestamp.from_datetime(start_datetime)
|
|
|
|
|
end_timestamp = DatabaseTimestamp.from_datetime(end_datetime)
|
|
|
|
|
|
|
|
|
|
start_previous_timestamp = sequence.db_previous_timestamp(start_timestamp)
|
|
|
|
|
assert start_previous_timestamp == record1_timestamp
|
|
|
|
|
end_next_timestamp = sequence.db_next_timestamp(end_timestamp)
|
|
|
|
|
assert end_next_timestamp == record2_timestamp
|
|
|
|
|
|
|
|
|
|
# Test
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=start_datetime,
|
|
|
|
|
end_datetime=end_datetime,
|
|
|
|
|
interval=interval,
|
|
|
|
|
fill_method="linear",
|
|
|
|
|
boundary="context",
|
|
|
|
|
)
|
|
|
|
|
np.testing.assert_equal(array, [1.5])
|
|
|
|
|
|
2024-12-29 18:42:49 +01:00
|
|
|
def test_key_to_array_ffill(self, sequence):
|
|
|
|
|
"""Test key_to_array with forward filling for missing values."""
|
|
|
|
|
interval = to_duration("1 hour")
|
|
|
|
|
record1 = self.create_test_record(pendulum.datetime(2023, 11, 6, 0), 0.8)
|
|
|
|
|
record2 = self.create_test_record(pendulum.datetime(2023, 11, 6, 2), 1.0)
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=pendulum.datetime(2023, 11, 6),
|
|
|
|
|
end_datetime=pendulum.datetime(2023, 11, 6, 3),
|
|
|
|
|
interval=interval,
|
|
|
|
|
fill_method="ffill",
|
|
|
|
|
)
|
|
|
|
|
assert len(array) == 3
|
|
|
|
|
assert array[0] == 0.8
|
|
|
|
|
assert array[1] == 0.8 # Forward-filled value
|
|
|
|
|
assert array[2] == 1.0
|
|
|
|
|
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
def test_key_to_array_ffill_one_value(self, sequence):
|
|
|
|
|
"""Test key_to_array with forward filling for missing values and only one value at end available."""
|
|
|
|
|
interval = to_duration("1 hour")
|
|
|
|
|
record1 = self.create_test_record(pendulum.datetime(2023, 11, 6, 2), 1.0)
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=pendulum.datetime(2023, 11, 6),
|
|
|
|
|
end_datetime=pendulum.datetime(2023, 11, 6, 4),
|
|
|
|
|
interval=interval,
|
|
|
|
|
fill_method="ffill",
|
|
|
|
|
)
|
|
|
|
|
assert len(array) == 4
|
|
|
|
|
assert array[0] == 1.0 # Backward-filled value
|
|
|
|
|
assert array[1] == 1.0 # Backward-filled value
|
|
|
|
|
assert array[2] == 1.0
|
|
|
|
|
assert array[2] == 1.0 # Forward-filled value
|
|
|
|
|
|
2024-12-29 18:42:49 +01:00
|
|
|
def test_key_to_array_bfill(self, sequence):
|
|
|
|
|
"""Test key_to_array with backward filling for missing values."""
|
|
|
|
|
interval = to_duration("1 hour")
|
|
|
|
|
record1 = self.create_test_record(pendulum.datetime(2023, 11, 6, 0), 0.8)
|
|
|
|
|
record2 = self.create_test_record(pendulum.datetime(2023, 11, 6, 2), 1.0)
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=pendulum.datetime(2023, 11, 6),
|
|
|
|
|
end_datetime=pendulum.datetime(2023, 11, 6, 3),
|
|
|
|
|
interval=interval,
|
|
|
|
|
fill_method="bfill",
|
|
|
|
|
)
|
|
|
|
|
assert len(array) == 3
|
|
|
|
|
assert array[0] == 0.8
|
|
|
|
|
assert array[1] == 1.0 # Backward-filled value
|
|
|
|
|
assert array[2] == 1.0
|
|
|
|
|
|
|
|
|
|
def test_key_to_array_with_truncation(self, sequence):
|
|
|
|
|
"""Test truncation behavior in key_to_array."""
|
|
|
|
|
interval = to_duration("1 hour")
|
|
|
|
|
record1 = self.create_test_record(pendulum.datetime(2023, 11, 5, 23), 0.8)
|
|
|
|
|
record2 = self.create_test_record(pendulum.datetime(2023, 11, 6, 1), 1.0)
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
#assert sequence is None
|
|
|
|
|
|
2024-12-29 18:42:49 +01:00
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
start_datetime=pendulum.datetime(2023, 11, 5, 23),
|
2024-12-29 18:42:49 +01:00
|
|
|
end_datetime=pendulum.datetime(2023, 11, 6, 2),
|
|
|
|
|
interval=interval,
|
|
|
|
|
)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
|
|
|
|
|
assert len(array) == 3
|
|
|
|
|
assert array[0] == 0.8
|
|
|
|
|
assert array[1] == 0.9 # Interpolated from previous day
|
|
|
|
|
assert array[2] == 1.0
|
2024-12-29 18:42:49 +01:00
|
|
|
|
|
|
|
|
def test_key_to_array_with_none(self, sequence):
|
|
|
|
|
"""Test handling of empty series in key_to_array."""
|
|
|
|
|
interval = to_duration("1 hour")
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=pendulum.datetime(2023, 11, 6),
|
|
|
|
|
end_datetime=pendulum.datetime(2023, 11, 6, 3),
|
|
|
|
|
interval=interval,
|
|
|
|
|
)
|
|
|
|
|
assert isinstance(array, np.ndarray)
|
|
|
|
|
assert np.all(array == None)
|
|
|
|
|
|
|
|
|
|
def test_key_to_array_with_one(self, sequence):
|
|
|
|
|
"""Test handling of one element series in key_to_array."""
|
|
|
|
|
interval = to_duration("1 hour")
|
|
|
|
|
record1 = self.create_test_record(pendulum.datetime(2023, 11, 5, 23), 0.8)
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
start_datetime=pendulum.datetime(2023, 11, 5, 23),
|
2024-12-29 18:42:49 +01:00
|
|
|
end_datetime=pendulum.datetime(2023, 11, 6, 2),
|
|
|
|
|
interval=interval,
|
|
|
|
|
)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
assert len(array) == 3
|
|
|
|
|
assert array[0] == 0.8
|
|
|
|
|
assert array[1] == 0.8 # Interpolated from previous day
|
|
|
|
|
assert array[2] == 0.8 # Interpolated from previous day
|
2024-12-29 18:42:49 +01:00
|
|
|
|
|
|
|
|
def test_key_to_array_invalid_fill_method(self, sequence):
|
|
|
|
|
"""Test invalid fill_method raises an error."""
|
|
|
|
|
interval = to_duration("1 hour")
|
|
|
|
|
record1 = self.create_test_record(pendulum.datetime(2023, 11, 6, 0), 0.8)
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
|
|
|
|
|
with pytest.raises(ValueError, match="Unsupported fill method: invalid"):
|
|
|
|
|
sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=pendulum.datetime(2023, 11, 6),
|
|
|
|
|
end_datetime=pendulum.datetime(2023, 11, 6, 1),
|
|
|
|
|
interval=interval,
|
|
|
|
|
fill_method="invalid",
|
|
|
|
|
)
|
|
|
|
|
|
2025-12-30 22:08:21 +01:00
|
|
|
def test_key_to_array_resample_mean(self, sequence):
|
|
|
|
|
"""Test that numeric resampling uses mean when multiple values fall into one interval."""
|
|
|
|
|
interval = to_duration("1 hour")
|
|
|
|
|
# Insert values every 15 minutes within the same hour
|
|
|
|
|
record1 = self.create_test_record(pendulum.datetime(2023, 11, 6, 0, 0), 1.0)
|
|
|
|
|
record2 = self.create_test_record(pendulum.datetime(2023, 11, 6, 0, 15), 2.0)
|
|
|
|
|
record3 = self.create_test_record(pendulum.datetime(2023, 11, 6, 0, 30), 3.0)
|
|
|
|
|
record4 = self.create_test_record(pendulum.datetime(2023, 11, 6, 0, 45), 4.0)
|
|
|
|
|
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
sequence.insert_by_datetime(record3)
|
|
|
|
|
sequence.insert_by_datetime(record4)
|
|
|
|
|
|
|
|
|
|
# Resample to hourly interval, expecting the mean of the 4 values
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=pendulum.datetime(2023, 11, 6, 0),
|
|
|
|
|
end_datetime=pendulum.datetime(2023, 11, 6, 1),
|
|
|
|
|
interval=interval,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert isinstance(array, np.ndarray)
|
|
|
|
|
assert len(array) == 1 # one interval: 0:00-1:00
|
|
|
|
|
# The first interval mean = (1+2+3+4)/4 = 2.5
|
|
|
|
|
assert array[0] == pytest.approx(2.5)
|
|
|
|
|
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
# ------------------------------------------------------------------
|
|
|
|
|
# key_to_array — align_to_interval parameter
|
|
|
|
|
# ------------------------------------------------------------------
|
|
|
|
|
#
|
|
|
|
|
# The existing tests above use start_datetime values that already sit on
|
|
|
|
|
# clean hour/day boundaries, so the default alignment (origin=query_start)
|
|
|
|
|
# and clock alignment (origin=epoch-floor) produce identical results.
|
|
|
|
|
# The tests below specifically use off-boundary start times to expose
|
|
|
|
|
# the difference and verify the new parameter.
|
|
|
|
|
|
|
|
|
|
def test_key_to_array_align_false_origin_is_query_start(self, sequence):
|
|
|
|
|
"""Without align_to_interval the first bucket sits at query_start, not a clock boundary.
|
|
|
|
|
|
|
|
|
|
With start_datetime at 10:07:00 and 15-min interval the first resampled
|
|
|
|
|
bucket must be at 10:07:00 (origin = query_start), NOT at 10:00:00 or 10:15:00.
|
|
|
|
|
"""
|
|
|
|
|
# Off-boundary start: 10:07
|
|
|
|
|
start_dt = pendulum.datetime(2024, 6, 1, 10, 7, tz="UTC")
|
|
|
|
|
end_dt = pendulum.datetime(2024, 6, 1, 12, 7, tz="UTC")
|
|
|
|
|
|
|
|
|
|
# Records every 15 min so the resampled mean equals the input values
|
|
|
|
|
for m in range(0, 120, 15):
|
|
|
|
|
dt = pendulum.datetime(2024, 6, 1, 10, 7, tz="UTC").add(minutes=m)
|
|
|
|
|
sequence.insert_by_datetime(self.create_test_record(dt, float(m)))
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=start_dt,
|
|
|
|
|
end_datetime=end_dt,
|
|
|
|
|
interval=to_duration("15 minutes"),
|
|
|
|
|
fill_method="time",
|
|
|
|
|
boundary="strict",
|
|
|
|
|
align_to_interval=False,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert len(array) > 0
|
|
|
|
|
# Reconstruct the pandas index that key_to_array used: origin=start_dt
|
|
|
|
|
idx = pd.date_range(start=start_dt, periods=len(array), freq="900s")
|
|
|
|
|
# First bucket must be exactly at start_dt (10:07)
|
|
|
|
|
assert idx[0].minute == 7
|
|
|
|
|
assert idx[0].second == 0
|
|
|
|
|
|
|
|
|
|
def test_key_to_array_align_true_15min_buckets_on_quarter_hours(self, sequence):
|
|
|
|
|
"""align_to_interval=True produces timestamps on :00/:15/:30/:45 boundaries."""
|
|
|
|
|
# Off-boundary start: 10:07
|
|
|
|
|
start_dt = pendulum.datetime(2024, 6, 1, 10, 7, tz="UTC")
|
|
|
|
|
end_dt = pendulum.datetime(2024, 6, 1, 12, 7, tz="UTC")
|
|
|
|
|
|
|
|
|
|
# 1-min records across the window so resampling has data to work with
|
|
|
|
|
for m in range(0, 121):
|
|
|
|
|
dt = pendulum.datetime(2024, 6, 1, 10, 7, tz="UTC").add(minutes=m)
|
|
|
|
|
sequence.insert_by_datetime(self.create_test_record(dt, float(m)))
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=start_dt,
|
|
|
|
|
end_datetime=end_dt,
|
|
|
|
|
interval=to_duration("15 minutes"),
|
|
|
|
|
fill_method="time",
|
|
|
|
|
boundary="strict",
|
|
|
|
|
align_to_interval=True,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert len(array) > 0
|
|
|
|
|
# Reconstruct the epoch-aligned index that key_to_array must have used
|
|
|
|
|
import math
|
|
|
|
|
epoch = int(start_dt.timestamp())
|
|
|
|
|
floored_epoch = (epoch // 900) * 900 # floor to nearest 15-min boundary
|
|
|
|
|
idx = pd.date_range(
|
|
|
|
|
start=pd.Timestamp(floored_epoch, unit="s", tz="UTC"),
|
|
|
|
|
periods=len(array),
|
|
|
|
|
freq="900s",
|
|
|
|
|
)
|
|
|
|
|
# Every bucket must land on a :00/:15/:30/:45 minute mark with zero seconds
|
|
|
|
|
for ts in idx:
|
|
|
|
|
assert ts.minute % 15 == 0, (
|
|
|
|
|
f"Bucket at {ts} is not on a 15-min boundary (minute={ts.minute})"
|
|
|
|
|
)
|
|
|
|
|
assert ts.second == 0, (
|
|
|
|
|
f"Bucket at {ts} has non-zero seconds ({ts.second})"
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
def test_key_to_array_align_true_1hour_buckets_on_the_hour(self, sequence):
|
|
|
|
|
"""align_to_interval=True with 1-hour interval produces on-the-hour timestamps."""
|
|
|
|
|
# Off-boundary start: 10:23
|
|
|
|
|
start_dt = pendulum.datetime(2024, 6, 1, 10, 23, tz="UTC")
|
|
|
|
|
end_dt = pendulum.datetime(2024, 6, 1, 15, 23, tz="UTC")
|
|
|
|
|
|
|
|
|
|
for m in range(0, 301, 15):
|
|
|
|
|
dt = pendulum.datetime(2024, 6, 1, 10, 23, tz="UTC").add(minutes=m)
|
|
|
|
|
sequence.insert_by_datetime(self.create_test_record(dt, float(m)))
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=start_dt,
|
|
|
|
|
end_datetime=end_dt,
|
|
|
|
|
interval=to_duration("1 hour"),
|
|
|
|
|
fill_method="time",
|
|
|
|
|
boundary="strict",
|
|
|
|
|
align_to_interval=True,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert len(array) > 0
|
|
|
|
|
epoch = int(start_dt.timestamp())
|
|
|
|
|
floored_epoch = (epoch // 3600) * 3600 # floor to nearest hour
|
|
|
|
|
idx = pd.date_range(
|
|
|
|
|
start=pd.Timestamp(floored_epoch, unit="s", tz="UTC"),
|
|
|
|
|
periods=len(array),
|
|
|
|
|
freq="1h",
|
|
|
|
|
)
|
|
|
|
|
for ts in idx:
|
|
|
|
|
assert ts.minute == 0, (
|
|
|
|
|
f"Bucket at {ts} should be on the hour (minute={ts.minute})"
|
|
|
|
|
)
|
|
|
|
|
assert ts.second == 0, (
|
|
|
|
|
f"Bucket at {ts} has non-zero seconds ({ts.second})"
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
def test_key_to_array_align_true_when_start_already_on_boundary(self, sequence):
|
|
|
|
|
"""align_to_interval=True is a no-op when start_datetime is exactly on a boundary.
|
|
|
|
|
|
|
|
|
|
With start at a clean 15-min mark both modes must produce identical arrays.
|
|
|
|
|
"""
|
|
|
|
|
# Exactly on boundary: 10:00:00
|
|
|
|
|
start_dt = pendulum.datetime(2024, 6, 1, 10, 0, tz="UTC")
|
|
|
|
|
end_dt = pendulum.datetime(2024, 6, 1, 12, 0, tz="UTC")
|
|
|
|
|
|
|
|
|
|
for m in range(0, 121, 15):
|
|
|
|
|
dt = pendulum.datetime(2024, 6, 1, 10, 0, tz="UTC").add(minutes=m)
|
|
|
|
|
sequence.insert_by_datetime(self.create_test_record(dt, float(m)))
|
|
|
|
|
|
|
|
|
|
arr_aligned = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=start_dt,
|
|
|
|
|
end_datetime=end_dt,
|
|
|
|
|
interval=to_duration("15 minutes"),
|
|
|
|
|
fill_method="time",
|
|
|
|
|
boundary="strict",
|
|
|
|
|
align_to_interval=True,
|
|
|
|
|
)
|
|
|
|
|
arr_default = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=start_dt,
|
|
|
|
|
end_datetime=end_dt,
|
|
|
|
|
interval=to_duration("15 minutes"),
|
|
|
|
|
fill_method="time",
|
|
|
|
|
boundary="strict",
|
|
|
|
|
align_to_interval=False,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert len(arr_aligned) == len(arr_default)
|
|
|
|
|
np.testing.assert_array_almost_equal(arr_aligned, arr_default, decimal=6)
|
|
|
|
|
|
|
|
|
|
def test_key_to_array_align_true_without_start_datetime(self, sequence):
|
|
|
|
|
"""align_to_interval=True with no start_datetime must not raise.
|
|
|
|
|
|
|
|
|
|
Without a query_start there is no origin to snap; behaviour falls back
|
|
|
|
|
to 'start_day' (same as default). No exception is expected.
|
|
|
|
|
"""
|
|
|
|
|
for m in range(0, 121, 15):
|
|
|
|
|
dt = pendulum.datetime(2024, 6, 1, 10, 7, tz="UTC").add(minutes=m)
|
|
|
|
|
sequence.insert_by_datetime(self.create_test_record(dt, float(m)))
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=None,
|
|
|
|
|
end_datetime=pendulum.datetime(2024, 6, 1, 12, 7, tz="UTC"),
|
|
|
|
|
interval=to_duration("15 minutes"),
|
|
|
|
|
fill_method="time",
|
|
|
|
|
boundary="strict",
|
|
|
|
|
align_to_interval=True,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert isinstance(array, np.ndarray)
|
|
|
|
|
assert len(array) > 0
|
|
|
|
|
|
|
|
|
|
def test_key_to_array_align_true_output_within_requested_window(self, sequence):
|
|
|
|
|
"""align_to_interval=True truncates output to [start_datetime, end_datetime).
|
|
|
|
|
|
|
|
|
|
The epoch-floor origin may generate a bucket before start_datetime (e.g. 10:00
|
|
|
|
|
when start is 10:07), but key_to_array must truncate it away. The surviving
|
|
|
|
|
buckets are verified directly by reconstructing the index from the first
|
|
|
|
|
surviving timestamp (the first epoch-aligned bucket >= start_datetime).
|
|
|
|
|
|
|
|
|
|
Also checks that all surviving buckets are on 15-min clock boundaries.
|
|
|
|
|
"""
|
|
|
|
|
start_dt = pendulum.datetime(2024, 6, 1, 10, 7, tz="UTC")
|
|
|
|
|
end_dt = pendulum.datetime(2024, 6, 1, 13, 7, tz="UTC")
|
|
|
|
|
|
|
|
|
|
for m in range(0, 181):
|
|
|
|
|
dt = pendulum.datetime(2024, 6, 1, 10, 7, tz="UTC").add(minutes=m)
|
|
|
|
|
sequence.insert_by_datetime(self.create_test_record(dt, float(m)))
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=start_dt,
|
|
|
|
|
end_datetime=end_dt,
|
|
|
|
|
interval=to_duration("15 minutes"),
|
|
|
|
|
fill_method="time",
|
|
|
|
|
boundary="strict",
|
|
|
|
|
align_to_interval=True,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert len(array) > 0
|
|
|
|
|
|
|
|
|
|
# The first surviving bucket is the first epoch-aligned timestamp >= start_dt.
|
|
|
|
|
# Compute it the same way key_to_array does: floor then step forward if needed.
|
|
|
|
|
epoch = int(start_dt.timestamp())
|
|
|
|
|
floored_epoch = (epoch // 900) * 900
|
|
|
|
|
first_bucket = pd.Timestamp(floored_epoch, unit="s", tz="UTC")
|
|
|
|
|
if first_bucket < pd.Timestamp(start_dt):
|
|
|
|
|
first_bucket += pd.Timedelta(seconds=900)
|
|
|
|
|
|
|
|
|
|
idx = pd.date_range(start=first_bucket, periods=len(array), freq="900s")
|
|
|
|
|
|
|
|
|
|
start_pd = pd.Timestamp(start_dt)
|
|
|
|
|
end_pd = pd.Timestamp(end_dt)
|
|
|
|
|
for ts in idx:
|
|
|
|
|
assert ts >= start_pd, f"Bucket {ts} is before start_datetime {start_pd}"
|
|
|
|
|
assert ts < end_pd, f"Bucket {ts} is at or after end_datetime {end_pd}"
|
|
|
|
|
assert ts.minute % 15 == 0, f"Bucket {ts} is not on a 15-min boundary"
|
|
|
|
|
assert ts.second == 0, f"Bucket {ts} has non-zero seconds"
|
|
|
|
|
|
|
|
|
|
def test_key_to_array_align_true_preserves_mean_values(self, sequence):
|
|
|
|
|
"""align_to_interval=True does not corrupt resampled values.
|
|
|
|
|
|
|
|
|
|
A constant-valued series must resample to the same constant regardless
|
|
|
|
|
of bucket alignment.
|
|
|
|
|
"""
|
|
|
|
|
# 1-min records with constant value 42.0, starting off-boundary
|
|
|
|
|
start_dt = pendulum.datetime(2024, 6, 1, 10, 7, tz="UTC")
|
|
|
|
|
end_dt = pendulum.datetime(2024, 6, 1, 12, 7, tz="UTC")
|
|
|
|
|
|
|
|
|
|
for m in range(0, 121):
|
|
|
|
|
dt = pendulum.datetime(2024, 6, 1, 10, 7, tz="UTC").add(minutes=m)
|
|
|
|
|
sequence.insert_by_datetime(self.create_test_record(dt, 42.0))
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=start_dt,
|
|
|
|
|
end_datetime=end_dt,
|
|
|
|
|
interval=to_duration("15 minutes"),
|
|
|
|
|
fill_method="time",
|
|
|
|
|
boundary="strict",
|
|
|
|
|
align_to_interval=True,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert len(array) > 0
|
|
|
|
|
for v in array:
|
|
|
|
|
if v is not None:
|
|
|
|
|
assert abs(v - 42.0) < 1e-6, f"Expected 42.0, got {v}"
|
|
|
|
|
|
|
|
|
|
def test_key_to_array_align_true_compaction_call_pattern(self, sequence):
|
|
|
|
|
"""Verify the call pattern used by _db_compact_tier produces clock-aligned timestamps.
|
|
|
|
|
|
|
|
|
|
_db_compact_tier calls key_to_array with boundary='strict', fill_method='time',
|
|
|
|
|
align_to_interval=True on a window whose start has arbitrary sub-second precision.
|
|
|
|
|
All output buckets must land on 15-min boundaries so that compacted records are
|
|
|
|
|
stored at predictable, human-readable timestamps.
|
|
|
|
|
"""
|
|
|
|
|
# Non-round base time: 08:43 — chosen to expose any origin-alignment bug
|
|
|
|
|
base_dt = pendulum.datetime(2024, 6, 1, 8, 43, tz="UTC")
|
|
|
|
|
window_end = pendulum.datetime(2024, 6, 1, 11, 43, tz="UTC")
|
|
|
|
|
|
|
|
|
|
for m in range(0, 181):
|
|
|
|
|
dt = base_dt.add(minutes=m)
|
|
|
|
|
sequence.insert_by_datetime(self.create_test_record(dt, float(m)))
|
|
|
|
|
|
|
|
|
|
array = sequence.key_to_array(
|
|
|
|
|
key="data_value",
|
|
|
|
|
start_datetime=base_dt,
|
|
|
|
|
end_datetime=window_end,
|
|
|
|
|
interval=to_duration("15 minutes"),
|
|
|
|
|
fill_method="time",
|
|
|
|
|
boundary="strict",
|
|
|
|
|
align_to_interval=True,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert len(array) > 0
|
|
|
|
|
epoch = int(base_dt.timestamp())
|
|
|
|
|
floored_epoch = (epoch // 900) * 900
|
|
|
|
|
idx = pd.date_range(
|
|
|
|
|
start=pd.Timestamp(floored_epoch, unit="s", tz="UTC"),
|
|
|
|
|
periods=len(array),
|
|
|
|
|
freq="900s",
|
|
|
|
|
)
|
|
|
|
|
for ts in idx:
|
|
|
|
|
assert ts.minute % 15 == 0, (
|
|
|
|
|
f"Compacted record at {ts} is not on a 15-min boundary (minute={ts.minute})"
|
|
|
|
|
)
|
|
|
|
|
assert ts.second == 0, (
|
|
|
|
|
f"Compacted record at {ts} has non-zero seconds ({ts.second})"
|
|
|
|
|
)
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_delete_by_datetime_range(self, sequence):
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
dt1 = to_datetime("2023-11-05")
|
|
|
|
|
dt2 = to_datetime("2023-11-06")
|
|
|
|
|
dt3 = to_datetime("2023-11-07")
|
|
|
|
|
record1 = self.create_test_record(dt1, 0.8)
|
|
|
|
|
record2 = self.create_test_record(dt2, 0.9)
|
|
|
|
|
record3 = self.create_test_record(dt3, 1.0)
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
sequence.insert_by_datetime(record3)
|
2024-12-15 14:40:03 +01:00
|
|
|
assert len(sequence) == 3
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.delete_by_datetime(start_datetime=dt2, end_datetime=dt3)
|
2024-12-15 14:40:03 +01:00
|
|
|
assert len(sequence) == 2
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
assert sequence.records[0].date_time == dt1
|
|
|
|
|
assert sequence.records[1].date_time == dt3
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_delete_by_datetime_start(self, sequence):
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
dt1 = to_datetime("2023-11-05")
|
|
|
|
|
dt2 = to_datetime("2023-11-06")
|
|
|
|
|
record1 = self.create_test_record(dt1, 0.8)
|
|
|
|
|
record2 = self.create_test_record(dt2, 0.9)
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
2024-12-15 14:40:03 +01:00
|
|
|
assert len(sequence) == 2
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.delete_by_datetime(start_datetime=dt2)
|
2024-12-15 14:40:03 +01:00
|
|
|
assert len(sequence) == 1
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
assert sequence.records[0].date_time == dt1
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_delete_by_datetime_end(self, sequence):
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
dt1 = to_datetime("2023-11-05")
|
|
|
|
|
dt2 = to_datetime("2023-11-06")
|
|
|
|
|
record1 = self.create_test_record(dt1, 0.8)
|
|
|
|
|
record2 = self.create_test_record(dt2, 0.9)
|
|
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
2024-12-15 14:40:03 +01:00
|
|
|
assert len(sequence) == 2
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.delete_by_datetime(end_datetime=dt2)
|
2024-12-15 14:40:03 +01:00
|
|
|
assert len(sequence) == 1
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
assert sequence.records[0].date_time == dt2
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_to_dict(self, sequence):
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
dt = to_datetime("2023-11-06")
|
|
|
|
|
record = self.create_test_record(dt, 0.8)
|
|
|
|
|
sequence.insert_by_datetime(record)
|
2024-12-15 14:40:03 +01:00
|
|
|
data_dict = sequence.to_dict()
|
|
|
|
|
assert isinstance(data_dict, dict)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
# We need a new class - Sequences are singletons
|
|
|
|
|
sequence2 = DerivedSequence2.from_dict(data_dict)
|
|
|
|
|
assert sequence2.model_dump() == sequence.model_dump()
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_to_json(self, sequence):
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
dt = to_datetime("2023-11-06")
|
|
|
|
|
record = self.create_test_record(dt, 0.8)
|
|
|
|
|
sequence.insert_by_datetime(record)
|
2024-12-15 14:40:03 +01:00
|
|
|
json_str = sequence.to_json()
|
|
|
|
|
assert isinstance(json_str, str)
|
|
|
|
|
assert "2023-11-06" in json_str
|
2025-01-15 00:54:45 +01:00
|
|
|
assert ": 0.8" in json_str
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_from_json(self, sequence, sequence2):
|
|
|
|
|
json_str = sequence2.to_json()
|
|
|
|
|
sequence = sequence.from_json(json_str)
|
|
|
|
|
assert len(sequence) == len(sequence2)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
assert sequence.records[0].date_time == sequence2.records[0].date_time
|
|
|
|
|
assert sequence.records[0].data_value == sequence2.records[0].data_value
|
2024-12-15 14:40:03 +01:00
|
|
|
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
def test_key_to_value_exact_match(self, sequence):
|
|
|
|
|
"""Test key_to_value returns exact match when datetime matches a record."""
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
dt = to_datetime("2023-11-05")
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
record = self.create_test_record(dt, 0.75)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record)
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
result = sequence.key_to_value("data_value", dt)
|
|
|
|
|
assert result == 0.75
|
|
|
|
|
|
|
|
|
|
def test_key_to_value_nearest(self, sequence):
|
|
|
|
|
"""Test key_to_value returns value closest in time to the given datetime."""
|
|
|
|
|
record1 = self.create_test_record(datetime(2023, 11, 5, 12), 0.6)
|
|
|
|
|
record2 = self.create_test_record(datetime(2023, 11, 6, 12), 0.9)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
dt = datetime(2023, 11, 6, 10) # closer to record2
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
result = sequence.key_to_value("data_value", dt, time_window=to_duration("48 hours"))
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
assert result == 0.9
|
|
|
|
|
|
|
|
|
|
def test_key_to_value_nearest_after(self, sequence):
|
|
|
|
|
"""Test key_to_value returns value nearest after the given datetime."""
|
|
|
|
|
record1 = self.create_test_record(datetime(2023, 11, 5, 10), 0.7)
|
|
|
|
|
record2 = self.create_test_record(datetime(2023, 11, 5, 15), 0.8)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
dt = datetime(2023, 11, 5, 14) # closer to record2
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
result = sequence.key_to_value("data_value", dt, time_window=to_duration("48 hours"))
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
assert result == 0.8
|
|
|
|
|
|
|
|
|
|
def test_key_to_value_empty_sequence(self, sequence):
|
|
|
|
|
"""Test key_to_value returns None when sequence is empty."""
|
|
|
|
|
result = sequence.key_to_value("data_value", datetime(2023, 11, 5))
|
|
|
|
|
assert result is None
|
|
|
|
|
|
|
|
|
|
def test_key_to_value_missing_key(self, sequence):
|
|
|
|
|
"""Test key_to_value returns None when key is missing in records."""
|
|
|
|
|
record = self.create_test_record(datetime(2023, 11, 5), None)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record)
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
result = sequence.key_to_value("data_value", datetime(2023, 11, 5))
|
|
|
|
|
assert result is None
|
|
|
|
|
|
|
|
|
|
def test_key_to_value_multiple_records_with_none(self, sequence):
|
|
|
|
|
"""Test key_to_value skips records with None values."""
|
|
|
|
|
r1 = self.create_test_record(datetime(2023, 11, 5), None)
|
|
|
|
|
r2 = self.create_test_record(datetime(2023, 11, 6), 1.0)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(r1)
|
|
|
|
|
sequence.insert_by_datetime(r2)
|
|
|
|
|
result = sequence.key_to_value("data_value", datetime(2023, 11, 5, 12), time_window=to_duration("48 hours"))
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
assert result == 1.0
|
|
|
|
|
|
2024-12-15 14:40:03 +01:00
|
|
|
def test_key_to_dict(self, sequence):
|
|
|
|
|
record1 = self.create_test_record(datetime(2023, 11, 5), 0.8)
|
|
|
|
|
record2 = self.create_test_record(datetime(2023, 11, 6), 0.9)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
2024-12-15 14:40:03 +01:00
|
|
|
data_dict = sequence.key_to_dict("data_value")
|
|
|
|
|
assert isinstance(data_dict, dict)
|
|
|
|
|
assert data_dict[to_datetime(datetime(2023, 11, 5), as_string=True)] == 0.8
|
|
|
|
|
assert data_dict[to_datetime(datetime(2023, 11, 6), as_string=True)] == 0.9
|
|
|
|
|
|
|
|
|
|
def test_key_to_lists(self, sequence):
|
|
|
|
|
record1 = self.create_test_record(datetime(2023, 11, 5), 0.8)
|
|
|
|
|
record2 = self.create_test_record(datetime(2023, 11, 6), 0.9)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
2024-12-15 14:40:03 +01:00
|
|
|
dates, values = sequence.key_to_lists("data_value")
|
|
|
|
|
assert dates == [to_datetime(datetime(2023, 11, 5)), to_datetime(datetime(2023, 11, 6))]
|
|
|
|
|
assert values == [0.8, 0.9]
|
|
|
|
|
|
2025-02-12 21:35:51 +01:00
|
|
|
def test_to_dataframe_full_data(self, sequence):
|
|
|
|
|
"""Test conversion of all records to a DataFrame without filtering."""
|
|
|
|
|
record1 = self.create_test_record("2024-01-01T12:00:00Z", 10)
|
|
|
|
|
record2 = self.create_test_record("2024-01-01T13:00:00Z", 20)
|
|
|
|
|
record3 = self.create_test_record("2024-01-01T14:00:00Z", 30)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
sequence.insert_by_datetime(record3)
|
2025-02-12 21:35:51 +01:00
|
|
|
|
|
|
|
|
df = sequence.to_dataframe()
|
|
|
|
|
|
|
|
|
|
# Validate DataFrame structure
|
|
|
|
|
assert isinstance(df, pd.DataFrame)
|
|
|
|
|
assert not df.empty
|
|
|
|
|
assert len(df) == 3 # All records should be included
|
|
|
|
|
assert "data_value" in df.columns
|
|
|
|
|
|
|
|
|
|
def test_to_dataframe_with_filter(self, sequence):
|
|
|
|
|
"""Test filtering records by datetime range."""
|
|
|
|
|
record1 = self.create_test_record("2024-01-01T12:00:00Z", 10)
|
|
|
|
|
record2 = self.create_test_record("2024-01-01T13:00:00Z", 20)
|
|
|
|
|
record3 = self.create_test_record("2024-01-01T14:00:00Z", 30)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
sequence.insert_by_datetime(record3)
|
2025-02-12 21:35:51 +01:00
|
|
|
|
|
|
|
|
start = to_datetime("2024-01-01T12:30:00Z")
|
|
|
|
|
end = to_datetime("2024-01-01T14:00:00Z")
|
|
|
|
|
|
|
|
|
|
df = sequence.to_dataframe(start_datetime=start, end_datetime=end)
|
|
|
|
|
|
|
|
|
|
assert isinstance(df, pd.DataFrame)
|
|
|
|
|
assert not df.empty
|
|
|
|
|
assert len(df) == 1 # Only one record should match the range
|
|
|
|
|
assert df.index[0] == pd.Timestamp("2024-01-01T13:00:00Z")
|
|
|
|
|
|
|
|
|
|
def test_to_dataframe_no_matching_records(self, sequence):
|
|
|
|
|
"""Test when no records match the given datetime filter."""
|
|
|
|
|
record1 = self.create_test_record("2024-01-01T12:00:00Z", 10)
|
|
|
|
|
record2 = self.create_test_record("2024-01-01T13:00:00Z", 20)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
2025-02-12 21:35:51 +01:00
|
|
|
|
|
|
|
|
start = to_datetime("2024-01-01T14:00:00Z") # Start time after all records
|
|
|
|
|
end = to_datetime("2024-01-01T15:00:00Z")
|
|
|
|
|
|
|
|
|
|
df = sequence.to_dataframe(start_datetime=start, end_datetime=end)
|
|
|
|
|
|
|
|
|
|
assert isinstance(df, pd.DataFrame)
|
|
|
|
|
assert df.empty # No records should match
|
|
|
|
|
|
|
|
|
|
def test_to_dataframe_empty_sequence(self, sequence):
|
|
|
|
|
"""Test when DataSequence has no records."""
|
|
|
|
|
sequence = DataSequence(records=[])
|
|
|
|
|
|
|
|
|
|
df = sequence.to_dataframe()
|
|
|
|
|
|
|
|
|
|
assert isinstance(df, pd.DataFrame)
|
|
|
|
|
assert df.empty # Should return an empty DataFrame
|
|
|
|
|
|
|
|
|
|
def test_to_dataframe_no_start_datetime(self, sequence):
|
|
|
|
|
"""Test when only end_datetime is given (all past records should be included)."""
|
|
|
|
|
record1 = self.create_test_record("2024-01-01T12:00:00Z", 10)
|
|
|
|
|
record2 = self.create_test_record("2024-01-01T13:00:00Z", 20)
|
|
|
|
|
record3 = self.create_test_record("2024-01-01T14:00:00Z", 30)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
sequence.insert_by_datetime(record3)
|
2025-02-12 21:35:51 +01:00
|
|
|
|
|
|
|
|
end = to_datetime("2024-01-01T13:00:00Z") # Include only first record
|
|
|
|
|
|
|
|
|
|
df = sequence.to_dataframe(end_datetime=end)
|
|
|
|
|
|
|
|
|
|
assert isinstance(df, pd.DataFrame)
|
|
|
|
|
assert not df.empty
|
|
|
|
|
assert len(df) == 1
|
|
|
|
|
assert df.index[0] == pd.Timestamp("2024-01-01T12:00:00Z")
|
|
|
|
|
|
|
|
|
|
def test_to_dataframe_no_end_datetime(self, sequence):
|
|
|
|
|
"""Test when only start_datetime is given (all future records should be included)."""
|
|
|
|
|
record1 = self.create_test_record("2024-01-01T12:00:00Z", 10)
|
|
|
|
|
record2 = self.create_test_record("2024-01-01T13:00:00Z", 20)
|
|
|
|
|
record3 = self.create_test_record("2024-01-01T14:00:00Z", 30)
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
sequence.insert_by_datetime(record1)
|
|
|
|
|
sequence.insert_by_datetime(record2)
|
|
|
|
|
sequence.insert_by_datetime(record3)
|
2025-02-12 21:35:51 +01:00
|
|
|
|
|
|
|
|
start = to_datetime("2024-01-01T13:00:00Z") # Include last two records
|
|
|
|
|
|
|
|
|
|
df = sequence.to_dataframe(start_datetime=start)
|
|
|
|
|
|
|
|
|
|
assert isinstance(df, pd.DataFrame)
|
|
|
|
|
assert not df.empty
|
|
|
|
|
assert len(df) == 2
|
|
|
|
|
assert df.index[0] == pd.Timestamp("2024-01-01T13:00:00Z")
|
|
|
|
|
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
class TestDataProvider:
|
|
|
|
|
# Fixtures and helper functions
|
|
|
|
|
@pytest.fixture
|
|
|
|
|
def provider(self):
|
|
|
|
|
"""Fixture to provide an instance of TestDataProvider for testing."""
|
|
|
|
|
DerivedDataProvider.provider_enabled = True
|
|
|
|
|
DerivedDataProvider.provider_updated = False
|
|
|
|
|
return DerivedDataProvider()
|
|
|
|
|
|
|
|
|
|
@pytest.fixture
|
|
|
|
|
def sample_start_datetime(self):
|
|
|
|
|
"""Fixture for a sample start datetime."""
|
|
|
|
|
return to_datetime(datetime(2024, 11, 1, 12, 0))
|
|
|
|
|
|
|
|
|
|
def create_test_record(self, date, value):
|
|
|
|
|
"""Helper function to create a test DataRecord."""
|
|
|
|
|
return DerivedRecord(date_time=date, data_value=value)
|
|
|
|
|
|
|
|
|
|
# Tests
|
|
|
|
|
|
|
|
|
|
def test_singleton_behavior(self, provider):
|
|
|
|
|
"""Test that DataProvider enforces singleton behavior."""
|
|
|
|
|
instance1 = provider
|
|
|
|
|
instance2 = DerivedDataProvider()
|
2025-02-24 10:00:09 +01:00
|
|
|
assert instance1 is instance2, (
|
|
|
|
|
"Singleton pattern is not enforced; instances are not the same."
|
|
|
|
|
)
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_update_method_with_defaults(self, provider, sample_start_datetime, monkeypatch):
|
|
|
|
|
"""Test the `update` method with default parameters."""
|
|
|
|
|
ems_eos = get_ems()
|
|
|
|
|
|
|
|
|
|
ems_eos.set_start_datetime(sample_start_datetime)
|
|
|
|
|
provider.update_data()
|
|
|
|
|
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
assert provider.ems_start_datetime == sample_start_datetime
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_update_method_force_enable(self, provider, monkeypatch):
|
|
|
|
|
"""Test that `update` executes when `force_enable` is True, even if `enabled` is False."""
|
|
|
|
|
# Override enabled to return False for this test
|
|
|
|
|
DerivedDataProvider.provider_enabled = False
|
|
|
|
|
DerivedDataProvider.provider_updated = False
|
|
|
|
|
provider.update_data(force_enable=True)
|
|
|
|
|
assert provider.enabled() is False, "Provider should be disabled, but enabled() is True."
|
2025-02-24 10:00:09 +01:00
|
|
|
assert DerivedDataProvider.provider_updated is True, (
|
|
|
|
|
"Provider should have been executed, but was not."
|
|
|
|
|
)
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_delete_by_datetime(self, provider, sample_start_datetime):
|
|
|
|
|
"""Test `delete_by_datetime` method for removing records by datetime range."""
|
|
|
|
|
# Add records to the provider for deletion testing
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
records = [
|
2024-12-15 14:40:03 +01:00
|
|
|
self.create_test_record(sample_start_datetime - to_duration("3 hours"), 1),
|
|
|
|
|
self.create_test_record(sample_start_datetime - to_duration("1 hour"), 2),
|
|
|
|
|
self.create_test_record(sample_start_datetime + to_duration("1 hour"), 3),
|
|
|
|
|
]
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
for record in records:
|
|
|
|
|
provider.insert_by_datetime(record)
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
provider.delete_by_datetime(
|
|
|
|
|
start_datetime=sample_start_datetime - to_duration("2 hours"),
|
|
|
|
|
end_datetime=sample_start_datetime + to_duration("2 hours"),
|
|
|
|
|
)
|
2025-02-24 10:00:09 +01:00
|
|
|
assert len(provider.records) == 1, (
|
|
|
|
|
"Only one record should remain after deletion by datetime."
|
|
|
|
|
)
|
|
|
|
|
assert provider.records[0].date_time == sample_start_datetime - to_duration("3 hours"), (
|
|
|
|
|
"Unexpected record remains."
|
|
|
|
|
)
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
class NewTestDataImportProvider:
|
|
|
|
|
|
2024-12-15 14:40:03 +01:00
|
|
|
# Fixtures and helper functions
|
|
|
|
|
@pytest.fixture
|
|
|
|
|
def provider(self):
|
|
|
|
|
"""Fixture to provide an instance of DerivedDataImportProvider for testing."""
|
|
|
|
|
DerivedDataImportProvider.provider_enabled = True
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
DerivedDataImportProvider.provider_updated = True
|
2024-12-15 14:40:03 +01:00
|
|
|
return DerivedDataImportProvider()
|
|
|
|
|
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
# import_from_dict
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
def test_import_from_dict_basic(self, provider):
|
|
|
|
|
data = {
|
|
|
|
|
"start_datetime": "2024-01-01 00:00:00",
|
|
|
|
|
"interval": "1 hour",
|
|
|
|
|
"power": [1, 2, 3],
|
|
|
|
|
}
|
2024-12-15 14:40:03 +01:00
|
|
|
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
provider.import_from_dict(data)
|
2024-12-15 14:40:03 +01:00
|
|
|
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
assert provider.records is not None
|
|
|
|
|
assert provider.records[0]["power"] == 1
|
|
|
|
|
assert provider.records[1]["power"] == 2
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
def test_import_from_dict_default_start_and_interval(self, provider):
|
|
|
|
|
data = {
|
|
|
|
|
"power": [10, 20],
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
provider.import_from_dict(data)
|
|
|
|
|
|
|
|
|
|
assert len(provider._updates) == 2
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_import_from_dict_with_prefix(self, provider):
|
|
|
|
|
data = {
|
|
|
|
|
"load_power": [1, 2],
|
|
|
|
|
"other": [5, 6],
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
provider.import_from_dict(data, key_prefix="load")
|
|
|
|
|
|
|
|
|
|
assert len(provider._updates) == 2
|
|
|
|
|
assert all(update[1] == "load_power" for update in provider._updates)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_import_from_dict_mismatching_lengths(self, provider):
|
|
|
|
|
data = {
|
|
|
|
|
"power": [1, 2],
|
|
|
|
|
"voltage": [1],
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
with pytest.raises(ValueError):
|
|
|
|
|
provider.import_from_dict(data)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_import_from_dict_invalid_interval(self, provider):
|
|
|
|
|
data = {
|
|
|
|
|
"interval": "17 minutes", # does not divide hour
|
|
|
|
|
"power": [1, 2, 3],
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
with pytest.raises(NotImplementedError):
|
|
|
|
|
provider.import_from_dict(data)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_import_from_dict_skips_none_and_nan(self, provider):
|
|
|
|
|
data = {
|
|
|
|
|
"power": [1, None, np.nan, 4],
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
provider.import_from_dict(data)
|
|
|
|
|
|
|
|
|
|
# only 1 and 4 should be written
|
|
|
|
|
assert len(provider._updates) == 2
|
|
|
|
|
assert provider._updates[0][2] == 1
|
|
|
|
|
assert provider._updates[1][2] == 4
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_import_from_dict_invalid_value_type(self, provider):
|
|
|
|
|
data = {
|
|
|
|
|
"power": "not a list"
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
with pytest.raises(ValueError):
|
|
|
|
|
provider.import_from_dict(data)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
# import_from_dataframe
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
def test_import_from_dataframe_with_datetime_index(self, provider):
|
|
|
|
|
index = pd.date_range("2024-01-01", periods=3, freq="H")
|
|
|
|
|
df = pd.DataFrame({"power": [1, 2, 3]}, index=index)
|
|
|
|
|
|
|
|
|
|
provider.import_from_dataframe(df)
|
|
|
|
|
|
|
|
|
|
assert len(provider._updates) == 3
|
|
|
|
|
assert provider._updates[0][2] == 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_import_from_dataframe_without_datetime_index(self, provider):
|
|
|
|
|
df = pd.DataFrame({"power": [5, 6, 7]})
|
|
|
|
|
|
|
|
|
|
provider.import_from_dataframe(
|
|
|
|
|
df,
|
|
|
|
|
start_datetime=datetime(2024, 1, 1),
|
|
|
|
|
interval=to_duration("1 hour"),
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
assert len(provider._updates) == 3
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_import_from_dataframe_prefix_filter(self, provider):
|
|
|
|
|
df = pd.DataFrame({
|
|
|
|
|
"load_power": [1, 2],
|
|
|
|
|
"other": [3, 4],
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
provider.import_from_dataframe(df, key_prefix="load")
|
|
|
|
|
|
|
|
|
|
assert len(provider._updates) == 2
|
|
|
|
|
assert all(update[1] == "load_power" for update in provider._updates)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_import_from_dataframe_invalid_input(self, provider):
|
|
|
|
|
with pytest.raises(ValueError):
|
|
|
|
|
provider.import_from_dataframe("not a dataframe")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
# import_from_json
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
def test_import_from_json_simple_dict(self, provider):
|
|
|
|
|
json_str = json.dumps({
|
|
|
|
|
"power": [1, 2, 3]
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
provider.import_from_json(json_str)
|
|
|
|
|
|
|
|
|
|
assert len(provider._updates) == 3
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def test_import_from_json_invalid(self, provider):
|
|
|
|
|
with pytest.raises(ValueError):
|
|
|
|
|
provider.import_from_json("this is not json")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
# import_from_file
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
def test_import_from_file(self, provider, tmp_path):
|
|
|
|
|
file_path = tmp_path / "data.json"
|
|
|
|
|
|
|
|
|
|
file_path.write_text(json.dumps({
|
|
|
|
|
"power": [1, 2]
|
|
|
|
|
}))
|
|
|
|
|
|
|
|
|
|
provider.import_from_file(file_path)
|
|
|
|
|
|
|
|
|
|
assert len(provider._updates) == 2
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class TestDataContainer:
|
|
|
|
|
# Fixture and helpers
|
|
|
|
|
@pytest.fixture
|
|
|
|
|
def container(self):
|
|
|
|
|
container = DerivedDataContainer()
|
|
|
|
|
return container
|
|
|
|
|
|
|
|
|
|
@pytest.fixture
|
|
|
|
|
def container_with_providers(self):
|
|
|
|
|
record1 = self.create_test_record(datetime(2023, 11, 5), 1)
|
|
|
|
|
record2 = self.create_test_record(datetime(2023, 11, 6), 2)
|
|
|
|
|
record3 = self.create_test_record(datetime(2023, 11, 7), 3)
|
|
|
|
|
provider = DerivedDataProvider()
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
provider.delete_by_datetime(start_datetime=None, end_datetime=None)
|
2024-12-15 14:40:03 +01:00
|
|
|
assert len(provider) == 0
|
Add database support for measurements and historic prediction data. (#848)
The database supports backend selection, compression, incremental data load,
automatic data saving to storage, automatic vaccum and compaction.
Make SQLite3 and LMDB database backends available.
Update tests for new interface conventions regarding data sequences,
data containers, data providers. This includes the measurements provider and
the prediction providers.
Add database documentation.
The fix includes several bug fixes that are not directly related to the database
implementation but are necessary to keep EOS running properly and to test and
document the changes.
* fix: config eos test setup
Make the config_eos fixture generate a new instance of the config_eos singleton.
Use correct env names to setup data folder path.
* fix: startup with no config
Make cache and measurements complain about missing data path configuration but
do not bail out.
* fix: soc data preparation and usage for genetic optimization.
Search for soc measurments 48 hours around the optimization start time.
Only clamp soc to maximum in battery device simulation.
* fix: dashboard bailout on zero value solution display
Do not use zero values to calculate the chart values adjustment for display.
* fix: openapi generation script
Make the script also replace data_folder_path and data_output_path to hide
real (test) environment pathes.
* feat: add make repeated task function
make_repeated_task allows to wrap a function to be repeated cyclically.
* chore: removed index based data sequence access
Index based data sequence access does not make sense as the sequence can be backed
by the database. The sequence is now purely time series data.
* chore: refactor eos startup to avoid module import startup
Avoid module import initialisation expecially of the EOS configuration.
Config mutation, singleton initialization, logging setup, argparse parsing,
background task definitions depending on config and environment-dependent behavior
is now done at function startup.
* chore: introduce retention manager
A single long-running background task that owns the scheduling of all periodic
server-maintenance jobs (cache cleanup, DB autosave, …)
* chore: canonicalize timezone name for UTC
Timezone names that are semantically identical to UTC are canonicalized to UTC.
* chore: extend config file migration for default value handling
Extend the config file migration handling values None or nonexisting values
that will invoke a default value generation in the new config file. Also
adapt test to handle this situation.
* chore: extend datetime util test cases
* chore: make version test check for untracked files
Check for files that are not tracked by git. Version calculation will be
wrong if these files will not be commited.
* chore: bump pandas to 3.0.0
Pandas 3.0 now performs inference on the appropriate resolution (a.k.a. unit)
for the output dtype which may become datetime64[us] (before it was ns). Also
numeric dtype detection is now more strict which needs a different detection for
numerics.
* chore: bump pydantic-settings to 2.12.0
pydantic-settings 2.12.0 under pytest creates a different behaviour. The tests
were adapted and a workaround was introduced. Also ConfigEOS was adapted
to allow for fine grain initialization control to be able to switch
off certain settings such as file settings during test.
* chore: remove sci learn kit from dependencies
The sci learn kit is not strictly necessary as long as we have scipy.
* chore: add documentation mode guarding for sphinx autosummary
Sphinx autosummary excecutes functions. Prevent exceptions in case of pure doc
mode.
* chore: adapt docker-build CI workflow to stricter GitHub handling
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2026-02-22 14:12:42 +01:00
|
|
|
provider.insert_by_datetime(record1)
|
|
|
|
|
provider.insert_by_datetime(record2)
|
|
|
|
|
provider.insert_by_datetime(record3)
|
2024-12-15 14:40:03 +01:00
|
|
|
assert len(provider) == 3
|
|
|
|
|
container = DerivedDataContainer()
|
|
|
|
|
container.providers.clear()
|
|
|
|
|
assert len(container.providers) == 0
|
|
|
|
|
container.providers.append(provider)
|
|
|
|
|
assert len(container.providers) == 1
|
|
|
|
|
return container
|
|
|
|
|
|
|
|
|
|
def create_test_record(self, date, value):
|
|
|
|
|
"""Helper function to create a test DataRecord."""
|
|
|
|
|
return DerivedRecord(date_time=date, data_value=value)
|
|
|
|
|
|
|
|
|
|
def test_append_provider(self, container):
|
|
|
|
|
assert len(container.providers) == 0
|
|
|
|
|
container.providers.append(DerivedDataProvider())
|
|
|
|
|
assert len(container.providers) == 1
|
|
|
|
|
assert isinstance(container.providers[0], DerivedDataProvider)
|
|
|
|
|
|
|
|
|
|
@pytest.mark.skip(reason="type check not implemented")
|
|
|
|
|
def test_append_provider_invalid_type(self, container):
|
|
|
|
|
with pytest.raises(ValueError, match="must be an instance of DataProvider"):
|
|
|
|
|
container.providers.append("not_a_provider")
|
|
|
|
|
|
|
|
|
|
def test_getitem_existing_key(self, container_with_providers):
|
|
|
|
|
assert len(container_with_providers.providers) == 1
|
|
|
|
|
# check all keys are available (don't care for position)
|
|
|
|
|
for key in ["data_value", "date_time"]:
|
|
|
|
|
assert key in list(container_with_providers.keys())
|
|
|
|
|
series = container_with_providers["data_value"]
|
|
|
|
|
assert isinstance(series, pd.Series)
|
|
|
|
|
assert series.name == "data_value"
|
|
|
|
|
assert series.tolist() == [1.0, 2.0, 3.0]
|
|
|
|
|
|
|
|
|
|
def test_getitem_non_existing_key(self, container_with_providers):
|
|
|
|
|
with pytest.raises(KeyError, match="No data found for key 'non_existent_key'"):
|
|
|
|
|
container_with_providers["non_existent_key"]
|
|
|
|
|
|
|
|
|
|
def test_setitem_existing_key(self, container_with_providers):
|
|
|
|
|
new_series = container_with_providers["data_value"]
|
|
|
|
|
new_series[:] = [4, 5, 6]
|
|
|
|
|
container_with_providers["data_value"] = new_series
|
|
|
|
|
series = container_with_providers["data_value"]
|
|
|
|
|
assert series.name == "data_value"
|
|
|
|
|
assert series.tolist() == [4, 5, 6]
|
|
|
|
|
|
|
|
|
|
def test_setitem_invalid_value(self, container_with_providers):
|
|
|
|
|
with pytest.raises(ValueError, match="Value must be an instance of pd.Series"):
|
|
|
|
|
container_with_providers["test_key"] = "not_a_series"
|
|
|
|
|
|
|
|
|
|
def test_setitem_non_existing_key(self, container_with_providers):
|
|
|
|
|
new_series = pd.Series([4, 5, 6], name="non_existent_key")
|
|
|
|
|
with pytest.raises(KeyError, match="Key 'non_existent_key' not found"):
|
|
|
|
|
container_with_providers["non_existent_key"] = new_series
|
|
|
|
|
|
|
|
|
|
def test_delitem_existing_key(self, container_with_providers):
|
|
|
|
|
del container_with_providers["data_value"]
|
|
|
|
|
series = container_with_providers["data_value"]
|
|
|
|
|
assert series.name == "data_value"
|
2024-12-29 18:42:49 +01:00
|
|
|
assert series.tolist() == []
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_delitem_non_existing_key(self, container_with_providers):
|
|
|
|
|
with pytest.raises(KeyError, match="Key 'non_existent_key' not found"):
|
|
|
|
|
del container_with_providers["non_existent_key"]
|
|
|
|
|
|
|
|
|
|
def test_len(self, container_with_providers):
|
fix: automatic optimization (#596)
This fix implements the long term goal to have the EOS server run optimization (or
energy management) on regular intervals automatically. Thus clients can request
the current energy management plan at any time and it is updated on regular
intervals without interaction by the client.
This fix started out to "only" make automatic optimization (or energy management)
runs working. It turned out there are several endpoints that in some way
update predictions or run the optimization. To lock against such concurrent attempts
the code had to be refactored to allow control of execution. During refactoring it
became clear that some classes and files are named without a proper reference
to their usage. Thus not only refactoring but also renaming became necessary.
The names are still not the best, but I hope they are more intuitive.
The fix includes several bug fixes that are not directly related to the automatic optimization
but are necessary to keep EOS running properly to do the automatic optimization and
to test and document the changes.
This is a breaking change as the configuration structure changed once again and
the server API was also enhanced and streamlined. The server API that is used by
Andreas and Jörg in their videos has not changed.
* fix: automatic optimization
Allow optimization to automatically run on configured intervals gathering all
optimization parameters from configuration and predictions. The automatic run
can be configured to only run prediction updates skipping the optimization.
Extend documentaion to also cover automatic optimization. Lock automatic runs
against runs initiated by the /optimize or other endpoints. Provide new
endpoints to retrieve the energy management plan and the genetic solution
of the latest automatic optimization run. Offload energy management to thread
pool executor to keep the app more responsive during the CPU heavy optimization
run.
* fix: EOS servers recognize environment variables on startup
Force initialisation of EOS configuration on server startup to assure
all sources of EOS configuration are properly set up and read. Adapt
server tests and configuration tests to also test for environment
variable configuration.
* fix: Remove 0.0.0.0 to localhost translation under Windows
EOS imposed a 0.0.0.0 to localhost translation under Windows for
convenience. This caused some trouble in user configurations. Now, as the
default IP address configuration is 127.0.0.1, the user is responsible
for to set up the correct Windows compliant IP address.
* fix: allow names for hosts additional to IP addresses
* fix: access pydantic model fields by class
Access by instance is deprecated.
* fix: down sampling key_to_array
* fix: make cache clear endpoint clear all cache files
Make /v1/admin/cache/clear clear all cache files. Before it only cleared
expired cache files by default. Add new endpoint /v1/admin/clear-expired
to only clear expired cache files.
* fix: timezonefinder returns Europe/Paris instead of Europe/Berlin
timezonefinder 8.10 got more inaccurate for timezones in europe as there is
a common timezone. Use new package tzfpy instead which is still returning
Europe/Berlin if you are in Germany. tzfpy also claims to be faster than
timezonefinder.
* fix: provider settings configuration
Provider configuration used to be a union holding the settings for several
providers. Pydantic union handling does not always find the correct type
for a provider setting. This led to exceptions in specific configurations.
Now provider settings are explicit comfiguration items for each possible
provider. This is a breaking change as the configuration structure was
changed.
* fix: ClearOutside weather prediction irradiance calculation
Pvlib needs a pandas time index. Convert time index.
* fix: test config file priority
Do not use config_eos fixture as this fixture already creates a config file.
* fix: optimization sample request documentation
Provide all data in documentation of optimization sample request.
* fix: gitlint blocking pip dependency resolution
Replace gitlint by commitizen. Gitlint is not actively maintained anymore.
Gitlint dependencies blocked pip from dependency resolution.
* fix: sync pre-commit config to actual dependency requirements
.pre-commit-config.yaml was out of sync, also requirements-dev.txt.
* fix: missing babel in requirements.txt
Add babel to requirements.txt
* feat: setup default device configuration for automatic optimization
In case the parameters for automatic optimization are not fully defined a
default configuration is setup to allow the automatic energy management
run. The default configuration may help the user to correctly define
the device configuration.
* feat: allow configuration of genetic algorithm parameters
The genetic algorithm parameters for number of individuals, number of
generations, the seed and penalty function parameters are now avaliable
as configuration options.
* feat: allow configuration of home appliance time windows
The time windows a home appliance is allowed to run are now configurable
by the configuration (for /v1 API) and also by the home appliance parameters
(for the classic /optimize API). If there is no such configuration the
time window defaults to optimization hours, which was the standard before
the change. Documentation on how to configure time windows is added.
* feat: standardize mesaurement keys for battery/ ev SoC measurements
The standardized measurement keys to report battery SoC to the device
simulations can now be retrieved from the device configuration as a
read-only config option.
* feat: feed in tariff prediction
Add feed in tarif predictions needed for automatic optimization. The feed in
tariff can be retrieved as fixed feed in tarif or can be imported. Also add
tests for the different feed in tariff providers. Extend documentation to
cover the feed in tariff providers.
* feat: add energy management plan based on S2 standard instructions
EOS can generate an energy management plan as a list of simple instructions.
May be retrieved by the /v1/energy-management/plan endpoint. The instructions
loosely follow the S2 energy management standard.
* feat: make measurement keys configurable by EOS configuration.
The fixed measurement keys are replaced by configurable measurement keys.
* feat: make pendulum DateTime, Date, Duration types usable for pydantic models
Use pydantic_extra_types.pendulum_dt to get pydantic pendulum types. Types are
added to the datetimeutil utility. Remove custom made pendulum adaptations
from EOS pydantic module. Make EOS modules use the pydantic pendulum types
managed by the datetimeutil module instead of the core pendulum types.
* feat: Add Time, TimeWindow, TimeWindowSequence and to_time to datetimeutil.
The time windows are are added to support home appliance time window
configuration. All time classes are also pydantic models. Time is the base
class for time definition derived from pendulum.Time.
* feat: Extend DataRecord by configurable field like data.
Configurable field like data was added to support the configuration of
measurement records.
* feat: Add additional information to health information
Version information is added to the health endpoints of eos and eosDash.
The start time of the last optimization and the latest run time of the energy
management is added to the EOS health information.
* feat: add pydantic merge model tests
* feat: add plan tab to EOSdash
The plan tab displays the current energy management instructions.
* feat: add predictions tab to EOSdash
The predictions tab displays the current predictions.
* feat: add cache management to EOSdash admin tab
The admin tab is extended by a section for cache management. It allows to
clear the cache.
* feat: add about tab to EOSdash
The about tab resembles the former hello tab and provides extra information.
* feat: Adapt changelog and prepare for release management
Release management using commitizen is added. The changelog file is adapted and
teh changelog and a description for release management is added in the
documentation.
* feat(doc): Improve install and devlopment documentation
Provide a more concise installation description in Readme.md and add extra
installation page and development page to documentation.
* chore: Use memory cache for interpolation instead of dict in inverter
Decorate calculate_self_consumption() with @cachemethod_until_update to cache
results in memory during an energy management/ optimization run. Replacement
of dict type caching in inverter is now possible because all optimization
runs are properly locked and the memory cache CacheUntilUpdateStore is properly
cleared at the start of any energy management/ optimization operation.
* chore: refactor genetic
Refactor the genetic algorithm modules for enhanced module structure and better
readability. Removed unnecessary and overcomplex devices singleton. Also
split devices configuration from genetic algorithm parameters to allow further
development independently from genetic algorithm parameter format. Move
charge rates configuration for electric vehicles from optimization to devices
configuration to allow to have different charge rates for different cars in
the future.
* chore: Rename memory cache to CacheEnergyManagementStore
The name better resembles the task of the cache to chache function and method
results for an energy management run. Also the decorator functions are renamed
accordingly: cachemethod_energy_management, cache_energy_management
* chore: use class properties for config/ems/prediction mixin classes
* chore: skip debug logs from mathplotlib
Mathplotlib is very noisy in debug mode.
* chore: automatically sync bokeh js to bokeh python package
bokeh was updated to 3.8.0, make JS CDN automatically follow the package version.
* chore: rename hello.py to about.py
Make hello.py the adapted EOSdash about page.
* chore: remove demo page from EOSdash
As no the plan and prediction pages are working without configuration, the demo
page is no longer necessary
* chore: split test_server.py for system test
Split test_server.py to create explicit test_system.py for system tests.
* chore: move doc utils to generate_config_md.py
The doc utils are only used in scripts/generate_config_md.py. Move it there to
attribute for strong cohesion.
* chore: improve pydantic merge model documentation
* chore: remove pendulum warning from readme
* chore: remove GitHub discussions from contributing documentation
Github discussions is to be replaced by Akkudoktor.net.
* chore(release): bump version to 0.1.0+dev for development
* build(deps): bump fastapi[standard] from 0.115.14 to 0.117.1
bump fastapi and make coverage version (for pytest-cov) explicit to avoid pip break.
* build(deps): bump uvicorn from 0.36.0 to 0.37.0
BREAKING CHANGE: EOS configuration changed. V1 API changed.
- The available_charge_rates_percent configuration is removed from optimization.
Use the new charge_rate configuration for the electric vehicle
- Optimization configuration parameter hours renamed to horizon_hours
- Device configuration now has to provide the number of devices and device
properties per device.
- Specific prediction provider configuration to be provided by explicit
configuration item (no union for all providers).
- Measurement keys to be provided as a list.
- New feed in tariff providers have to be configured.
- /v1/measurement/loadxxx endpoints are removed. Use generic mesaurement endpoints.
- /v1/admin/cache/clear now clears all cache files. Use
/v1/admin/cache/clear-expired to only clear all expired cache files.
Signed-off-by: Bobby Noelte <b0661n0e17e@gmail.com>
2025-10-28 02:50:31 +01:00
|
|
|
assert len(container_with_providers) == 5
|
2024-12-15 14:40:03 +01:00
|
|
|
|
|
|
|
|
def test_repr(self, container_with_providers):
|
|
|
|
|
representation = repr(container_with_providers)
|
|
|
|
|
assert representation.startswith("DerivedDataContainer(")
|
|
|
|
|
assert "DerivedDataProvider" in representation
|
|
|
|
|
|
|
|
|
|
def test_to_json(self, container_with_providers):
|
|
|
|
|
json_str = container_with_providers.to_json()
|
|
|
|
|
container_other = DerivedDataContainer.from_json(json_str)
|
|
|
|
|
assert container_other == container_with_providers
|
|
|
|
|
|
|
|
|
|
def test_from_json(self, container_with_providers):
|
|
|
|
|
json_str = container_with_providers.to_json()
|
|
|
|
|
container = DerivedDataContainer.from_json(json_str)
|
|
|
|
|
assert isinstance(container, DerivedDataContainer)
|
|
|
|
|
assert len(container.providers) == 1
|
|
|
|
|
assert container.providers[0] == container_with_providers.providers[0]
|
|
|
|
|
|
|
|
|
|
def test_provider_by_id(self, container_with_providers):
|
|
|
|
|
provider = container_with_providers.provider_by_id("DerivedDataProvider")
|
|
|
|
|
assert isinstance(provider, DerivedDataProvider)
|