fix: return only completed optimization results per run

src/akkudoktoreos/core/ems.py

@@ -180,7 +180,7 @@ class EnergyManagement(
         genetic0_seed: Optional[int] = None,
         force_enable: Optional[bool] = False,
         force_update: Optional[bool] = False,
-    ) -> None:
+    ) -> Optional[GeneticSolution | Genetic0Solution]:
         """Run the energy management.
 
         This method initializes the energy management run by setting its
@@ -224,7 +224,9 @@ class EnergyManagement(
                 even if a cached version is still valid.
 
         Returns:
-            None
+            The native optimization solution produced by this run, or None when
+            no optimization completed. Previous successful results remain available
+            through the solution getters after an aborted run.
         """
         async with EnergyManagement._run_lock:
             if mode is None:
@@ -234,7 +236,7 @@ class EnergyManagement(
                 raise ValueError(f"Unknown energy management mode {mode}.")
             if mode == EnergyManagementMode.DISABLED:
                 logger.info("Energy management run disabled.")
-                return
+                return None
 
             logger.info("Starting energy management run.")
 
@@ -277,7 +279,7 @@ class EnergyManagement(
             if mode == EnergyManagementMode.PREDICTION:
                 logger.info("Energy management run done (predictions updated)")
                 EnergyManagement._stage = EnergyManagementStage.IDLE
-                return
+                return None
 
             # --- Optimization ---
             EnergyManagement._stage = EnergyManagementStage.OPTIMIZATION
@@ -287,6 +289,8 @@ class EnergyManagement(
             if algorithm is None:
                 algorithm = self.config.optimization.algorithm
 
+            run_solution: GeneticSolution | Genetic0Solution
+
             # --- GENETIC algorithm ---
             if algorithm == OptimizationAlgorithm.GENETIC:
                 # Prepare optimization parameters
@@ -300,7 +304,7 @@ class EnergyManagement(
                             "Could not prepare optimisation parameters."
                         )
                         EnergyManagement._stage = EnergyManagementStage.IDLE
-                        return
+                        return None
 
                 # Take values from config if not given
                 if genetic_generations is None:
@@ -331,21 +335,20 @@ class EnergyManagement(
                         ),
                     )
 
+                    # Build all representations before publishing any of them.
+                    optimization_solution = await genetic_solution.optimization_solution()
+                    plan = genetic_solution.energy_management_plan()
+
                 except Exception:
                     logger.exception(f"{algorithm}: Energy management optimization failed.")
                     EnergyManagement._stage = EnergyManagementStage.IDLE
-                    return
+                    return None
 
-                # Make genetic solution public
+                # Publish a consistent set without yielding to another coroutine.
                 EnergyManagement._genetic_solution = genetic_solution
-
+                EnergyManagement._optimization_solution = optimization_solution
-                # Make optimization solution public
+                EnergyManagement._plan = plan
-                EnergyManagement._optimization_solution = (
+                run_solution = genetic_solution
-                    await genetic_solution.optimization_solution()
-                )
-
-                # Make plan public
-                EnergyManagement._plan = genetic_solution.energy_management_plan()
 
                 logger.debug(
                     "{}: Energy management genetic solution:\n{}",
@@ -366,7 +369,7 @@ class EnergyManagement(
                             "Could not prepare optimisation parameters."
                         )
                         EnergyManagement._stage = EnergyManagementStage.IDLE
-                        return
+                        return None
 
                 # Take values from config if not given
                 if genetic0_generations is None:
@@ -397,21 +400,20 @@ class EnergyManagement(
                         ),
                     )
 
+                    # Build all representations before publishing any of them.
+                    optimization_solution = await genetic0_solution.optimization_solution()
+                    plan = genetic0_solution.energy_management_plan()
+
                 except Exception:
                     logger.exception(f"{algorithm}: Energy management optimization failed.")
                     EnergyManagement._stage = EnergyManagementStage.IDLE
-                    return
+                    return None
 
-                # Make genetic0 solution public
+                # Publish a consistent set without yielding to another coroutine.
                 EnergyManagement._genetic0_solution = genetic0_solution
-
+                EnergyManagement._optimization_solution = optimization_solution
-                # Make optimization solution public
+                EnergyManagement._plan = plan
-                EnergyManagement._optimization_solution = (
+                run_solution = genetic0_solution
-                    await genetic0_solution.optimization_solution()
-                )
-
-                # Make plan public
-                EnergyManagement._plan = genetic0_solution.energy_management_plan()
 
                 logger.debug(
                     "{}: Energy management genetic solution:\n{}",
@@ -422,7 +424,7 @@ class EnergyManagement(
             else:
                 logger.error(f"Unknown optimization algorithm: '{algorithm}'. Skipping.")
                 EnergyManagement._stage = EnergyManagementStage.IDLE
-                return
+                return None
 
             optimization_duration = to_datetime() - optimization_start
             logger.info(
@@ -459,3 +461,4 @@ class EnergyManagement(
 
             # energy management run finished
             EnergyManagement._stage = EnergyManagementStage.IDLE
+            return run_solution

src/akkudoktoreos/server/eos.py

@@ -2197,7 +2197,7 @@ async def fastapi_optimize(
         start_datetime = to_datetime().set(hour=start_hour)
 
     # Ensure there is only one optimization/ energy management run at a time
-    await get_ems().run(
+    solution = await get_ems().run(
         start_datetime=start_datetime,
         mode=EnergyManagementMode.OPTIMIZATION,
         algorithm=OptimizationAlgorithm.GENETIC0,
@@ -2205,7 +2205,6 @@ async def fastapi_optimize(
         genetic0_generations=ngen,
     )
 
-    solution = get_ems().genetic0_solution()
     if solution is None:
         raise EOSProblem(
             status=404,

tests/test_optimize_run_result.py

@@ -0,0 +1,312 @@
+"""Offline regression tests for per-run optimization results and atomic publication."""
+
+from asyncio import Lock
+from types import SimpleNamespace
+from unittest.mock import AsyncMock, Mock
+
+import pytest
+
+from akkudoktoreos.core import ems as ems_module
+from akkudoktoreos.core.emsettings import EnergyManagementMode
+from akkudoktoreos.optimization.genetic0.genetic0params import (
+    Genetic0OptimizationParameters,
+)
+from akkudoktoreos.optimization.optimization import OptimizationAlgorithm
+from akkudoktoreos.utils.datetimeutil import to_datetime
+
+
+@pytest.fixture
+def offline_ems(monkeypatch):
+    """Run the actual EMS orchestration with no real adapters or prediction IO."""
+    cls = ems_module.EnergyManagement
+    for name in (
+        "_start_datetime",
+        "_last_run_datetime",
+        "_plan",
+        "_optimization_solution",
+        "_genetic_solution",
+        "_genetic0_solution",
+    ):
+        monkeypatch.setattr(cls, name, None)
+    monkeypatch.setattr(cls, "_stage", ems_module.EnergyManagementStage.IDLE)
+    monkeypatch.setattr(cls, "_run_lock", Lock())
+    monkeypatch.setattr(ems_module, "CacheEnergyManagementStore", Mock())
+    return SimpleNamespace(
+        config=SimpleNamespace(
+            ems=SimpleNamespace(mode=EnergyManagementMode.OPTIMIZATION),
+            optimization=SimpleNamespace(
+                algorithm=OptimizationAlgorithm.GENETIC,
+                genetic=SimpleNamespace(generations=3, seed=17),
+                genetic0=SimpleNamespace(generations=5, seed=29),
+            ),
+            server=SimpleNamespace(verbose=False),
+        ),
+        prediction=SimpleNamespace(update_data=AsyncMock()),
+        adapter=SimpleNamespace(update_data=AsyncMock()),
+        set_start_datetime=cls.set_start_datetime,
+    )
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("algorithm", list(OptimizationAlgorithm))
+@pytest.mark.parametrize("selection", ["configured", "explicit"])
+@pytest.mark.parametrize("supplied", [False, True])
+async def test_optimization_routes_only_selected_algorithm(
+    monkeypatch, offline_ems, algorithm, selection, supplied
+):
+    """Configuration selection and explicit overrides retain isolated async paths."""
+    selected_name = "Genetic" if algorithm == OptimizationAlgorithm.GENETIC else "Genetic0"
+    suffix = "genetic" if algorithm == OptimizationAlgorithm.GENETIC else "genetic0"
+    other_suffix = "genetic0" if suffix == "genetic" else "genetic"
+    sentinel_parameters = object()
+    sentinel_result, sentinel_plan = object(), object()
+    solution = SimpleNamespace(
+        optimization_solution=AsyncMock(return_value=sentinel_result),
+        energy_management_plan=Mock(return_value=sentinel_plan),
+    )
+    constructors = {}
+    preparers = {}
+    for prefix in ("Genetic", "Genetic0"):
+        constructor = Mock()
+        constructor.return_value.optimize_ems.return_value = solution
+        constructors[prefix] = constructor
+        monkeypatch.setattr(ems_module, prefix + "Optimization", constructor)
+        prepare = AsyncMock(return_value=sentinel_parameters)
+        preparers[prefix] = prepare
+        monkeypatch.setattr(
+            getattr(ems_module, prefix + "OptimizationParameters"), "prepare", prepare
+        )
+    kwargs = {"start_datetime": to_datetime("2026-09-16T10:00:00+02:00")}
+    if selection == "configured":
+        offline_ems.config.optimization.algorithm = algorithm
+    else:
+        offline_ems.config.optimization.algorithm = OptimizationAlgorithm(other_suffix.upper())
+        kwargs["algorithm"] = algorithm
+    if supplied:
+        kwargs[suffix + "_parameters"] = sentinel_parameters
+        kwargs[suffix + "_generations"] = 7
+        kwargs[suffix + "_seed"] = 43
+    run_result = await ems_module.EnergyManagement.run(offline_ems, **kwargs)
+    assert run_result is solution
+    selected = constructors[selected_name]
+    expected_config = getattr(offline_ems.config.optimization, suffix)
+    selected.assert_called_once_with(
+        verbose=False, fixed_seed=43 if supplied else expected_config.seed
+    )
+    selected.return_value.optimize_ems.assert_called_once_with(
+        start_hour=10,
+        parameters=sentinel_parameters,
+        ngen=7 if supplied else expected_config.generations,
+    )
+    other_name = "Genetic0" if selected_name == "Genetic" else "Genetic"
+    constructors[other_name].assert_not_called()
+    preparers[other_name].assert_not_awaited()
+    if supplied:
+        preparers[selected_name].assert_not_awaited()
+    else:
+        preparers[selected_name].assert_awaited_once_with()
+    solution.optimization_solution.assert_awaited_once_with()
+    solution.energy_management_plan.assert_called_once_with()
+    cls = ems_module.EnergyManagement
+    assert getattr(cls, "_" + suffix + "_solution") is solution
+    assert getattr(cls, "_" + other_suffix + "_solution") is None
+    assert cls.optimization_solution() is sentinel_result
+    assert cls.plan() is sentinel_plan
+    assert cls.stage() == ems_module.EnergyManagementStage.IDLE
+    offline_ems.prediction.update_data.assert_awaited_once_with(
+        force_enable=False, force_update=False
+    )
+    assert offline_ems.adapter.update_data.await_count == 2
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("mode", [EnergyManagementMode.DISABLED, EnergyManagementMode.PREDICTION])
+async def test_non_optimization_modes_never_optimize(monkeypatch, offline_ems, mode):
+    constructors = [Mock(), Mock()]
+    monkeypatch.setattr(ems_module, "GeneticOptimization", constructors[0])
+    monkeypatch.setattr(ems_module, "Genetic0Optimization", constructors[1])
+    offline_ems.config.ems.mode = mode
+    await ems_module.EnergyManagement.run(offline_ems)
+    for constructor in constructors:
+        constructor.assert_not_called()
+    assert offline_ems.prediction.update_data.await_count == (
+        mode == EnergyManagementMode.PREDICTION
+    )
+    assert offline_ems.adapter.update_data.await_count == (mode == EnergyManagementMode.PREDICTION)
+    assert ems_module.EnergyManagement.plan() is None
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("prefix", ["Genetic", "Genetic0"])
+async def test_missing_preparation_does_not_dispatch_controls(monkeypatch, offline_ems, prefix):
+    constructor = Mock()
+    monkeypatch.setattr(ems_module, prefix + "Optimization", constructor)
+    prepare = AsyncMock(return_value=None)
+    monkeypatch.setattr(getattr(ems_module, prefix + "OptimizationParameters"), "prepare", prepare)
+    await ems_module.EnergyManagement.run(
+        offline_ems, algorithm=OptimizationAlgorithm(prefix.upper())
+    )
+    prepare.assert_awaited_once_with()
+    constructor.assert_not_called()
+    assert offline_ems.adapter.update_data.await_count == 1  # acquisition only
+    assert ems_module.EnergyManagement.plan() is None
+    assert ems_module.EnergyManagement.stage() == ems_module.EnergyManagementStage.IDLE
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("start_hour", [None, 11])
+async def test_legacy_optimize_explicitly_uses_genetic0(monkeypatch, start_hour):
+    """Even with GENETIC configured, /optimize must never become the new optimizer."""
+    from akkudoktoreos.server import eos
+    from akkudoktoreos.server.rest.error import EOSProblem
+
+    fake = SimpleNamespace(
+        run=AsyncMock(return_value=None), genetic0_solution=Mock(return_value=None)
+    )
+    monkeypatch.setattr(eos, "get_ems", lambda: fake)
+    parameters = Genetic0OptimizationParameters.model_validate(
+        {
+            "ems": {
+                "pv_prognose_wh": [0.0, 100.0],
+                "gesamtlast": [100.0, 100.0],
+                "strompreis_euro_pro_wh": [0.0003, 0.0003],
+                "einspeiseverguetung_euro_pro_wh": 0.00008,
+                "preis_euro_pro_wh_akku": 0.0,
+            },
+            "pv_akku": None,
+            "eauto": None,
+            "inverter": None,
+        }
+    )
+    with pytest.raises(EOSProblem):
+        await eos.fastapi_optimize(parameters=parameters, start_hour=start_hour, ngen=2)
+    kwargs = fake.run.await_args.kwargs
+    assert kwargs["mode"] == EnergyManagementMode.OPTIMIZATION
+    assert kwargs["algorithm"] == OptimizationAlgorithm.GENETIC0
+    assert kwargs["genetic0_parameters"] is parameters
+    assert kwargs["genetic0_generations"] == 2
+    assert "genetic_parameters" not in kwargs
+    assert (
+        kwargs["start_datetime"] is None
+        if start_hour is None
+        else kwargs["start_datetime"].hour == 11
+    )
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("phase", ["optimizer", "conversion", "plan"])
+async def test_failed_legacy_http_run_never_reports_previous_solution(
+    monkeypatch, offline_ems, phase
+):
+    import json
+    from pathlib import Path
+    from types import MethodType
+
+    from httpx import ASGITransport, AsyncClient
+
+    from akkudoktoreos.optimization.genetic0.genetic0solution import Genetic0Solution
+    from akkudoktoreos.server import eos
+
+    cls = ems_module.EnergyManagement
+    data = json.loads(
+        (Path(__file__).parent / "testdata/genetic0/optimize_result_1.json").read_text()
+    )
+    previous = Genetic0Solution.model_validate(data)
+    monkeypatch.setattr(cls, "_genetic0_solution", previous)
+    constructor = Mock()
+    error = RuntimeError("synthetic " + phase + " failure")
+    native = SimpleNamespace(
+        optimization_solution=AsyncMock(return_value=object()),
+        energy_management_plan=Mock(return_value=object()),
+    )
+    constructor.return_value.optimize_ems.return_value = native
+    if phase == "optimizer":
+        constructor.return_value.optimize_ems.side_effect = error
+    elif phase == "conversion":
+        native.optimization_solution.side_effect = error
+    else:
+        native.energy_management_plan.side_effect = error
+    monkeypatch.setattr(ems_module, "Genetic0Optimization", constructor)
+    offline_ems.run = MethodType(cls.run, offline_ems)
+    offline_ems.genetic0_solution = cls.genetic0_solution
+    monkeypatch.setattr(eos, "get_ems", lambda: offline_ems)
+    async with AsyncClient(
+        transport=ASGITransport(app=eos.app, raise_app_exceptions=False), base_url="http://test"
+    ) as client:
+        response = await client.post("/optimize?ngen=1", json=data["parameters"])
+    constructor.return_value.optimize_ems.assert_called_once()
+    # This diagnostic proves a failure is the stale-success bug, not invalid input.
+    if response.status_code == 200:
+        assert response.json()["start_solution"] == previous.start_solution
+        assert response.json()["result"]["total_balance"] == previous.result.total_balance
+    assert response.status_code >= 400, (
+        "The failing optimizer returned HTTP 200 with the previous solution"
+    )
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("prefix", ["Genetic", "Genetic0"])
+@pytest.mark.parametrize("phase", ["conversion", "plan"])
+async def test_conversion_failure_preserves_consistent_previous_results(
+    monkeypatch, offline_ems, prefix, phase
+):
+    cls = ems_module.EnergyManagement
+    previous_specific, previous_generic, previous_plan = object(), object(), object()
+    suffix = prefix.lower()
+    monkeypatch.setattr(cls, "_" + suffix + "_solution", previous_specific)
+    monkeypatch.setattr(cls, "_optimization_solution", previous_generic)
+    monkeypatch.setattr(cls, "_plan", previous_plan)
+    conversion = AsyncMock(return_value=object())
+    solution = SimpleNamespace(optimization_solution=conversion, energy_management_plan=Mock())
+    if phase == "conversion":
+        conversion.side_effect = RuntimeError("synthetic conversion failure")
+    else:
+        solution.energy_management_plan.side_effect = RuntimeError("synthetic plan failure")
+    constructor = Mock()
+    constructor.return_value.optimize_ems.return_value = solution
+    monkeypatch.setattr(ems_module, prefix + "Optimization", constructor)
+    result = await cls.run(
+        offline_ems,
+        algorithm=OptimizationAlgorithm(prefix.upper()),
+        **{suffix + "_parameters": object()},
+    )
+    assert result is None
+    conversion.assert_awaited_once_with()
+    assert solution.energy_management_plan.call_count == (phase == "plan")
+    assert offline_ems.adapter.update_data.await_count == 1
+    assert (
+        getattr(cls, "_" + suffix + "_solution"),
+        cls.optimization_solution(),
+        cls.plan(),
+        cls.stage(),
+    ) == (
+        previous_specific,
+        previous_generic,
+        previous_plan,
+        ems_module.EnergyManagementStage.IDLE,
+    ), (
+        "Failed conversion published a new algorithm result beside the old plan and left EMS in OPTIMIZATION"
+    )
+
+
+@pytest.mark.asyncio
+async def test_legacy_endpoint_uses_run_return_value_not_last_cached_solution(monkeypatch):
+    import json
+    from pathlib import Path
+
+    from akkudoktoreos.optimization.genetic0.genetic0solution import Genetic0Solution
+    from akkudoktoreos.server import eos
+
+    data = json.loads(
+        (Path(__file__).parent / "testdata/genetic0/optimize_result_1.json").read_text()
+    )
+    produced = Genetic0Solution.model_validate(data)
+    fake = SimpleNamespace(
+        run=AsyncMock(return_value=produced),
+        genetic0_solution=Mock(side_effect=AssertionError("must use this run's result")),
+    )
+    monkeypatch.setattr(eos, "get_ems", lambda: fake)
+    result = await eos.fastapi_optimize(parameters=produced.parameters, ngen=1)
+    assert result.start_solution == produced.start_solution
+    assert result.result.total_balance == produced.result.total_balance
+    fake.genetic0_solution.assert_not_called()