Event-only modal cache + Settings edit gating + ES i18n polish

Modal caches now refresh on events only — the periodic prewarmer runs one-shot at startup, mount points split into static/runtime endpoints, and backups get a client 6-hour gate. Updates tab shows post-apply feedback and the script terminal no longer closes the parent modal. Settings adds edit gating on 3 cards with the 3-level contrast rule applied consistently. ES translation batch (~25 fixes) and What's New for 1.2.4.1-beta refreshed.
This commit is contained in:
MacRimi
2026-08-14 20:17:45 +02:00
parent 452902bd9a
commit 99ad69e8cf
19 changed files with 1317 additions and 490 deletions
+137 -78
View File
@@ -1616,11 +1616,30 @@ _vm_backups_cache: dict = {} # vmid -> (ts, payload)
_vm_apps_cache: dict = {} # vmid -> (ts, payload)
_vm_schedule_cache: dict = {} # vmid -> (ts, payload)
_vm_mounts_cache: dict = {} # vmid -> (ts, payload) — LXC only
_VM_DETAILS_TTL = 300 # config rarely changes without a user action
_VM_BACKUPS_TTL = 120 # storage scans — a new backup is an event
_VM_APPS_TTL = 600 # LXC apps register/unregister is rare
_VM_SCHEDULE_TTL = 900 # persisted schedule almost never changes
_VM_MOUNTS_TTL = 600 # mpX entries only change on manual edit
# Effective TTL is "indefinite": these caches are refreshed only
# by explicit event-based invalidation (`_vm_cache_invalidate` calls
# on start/stop/reboot, add/edit/delete app, apply update, edit
# schedule, create backup). No periodic poll — the prewarmer runs
# once at startup and then stays quiet. The rationale is that all
# of these payloads are backed by files (.conf / sidecar JSON /
# storage inventory) that only change through actions the Monitor
# either performs itself (invalidates in-line) or that require a
# guest restart (start/stop invalidates too). Runtime data that DOES
# change without an invalidation event lives in a different path:
# - Live CPU/mem/disk/network per guest → served by /api/vms (SWR
# poll every 2.5 s from the client, no cache here).
# - Firewall log → served on-demand, no cache at all.
# - Mount points runtime (df/stat/ad-hoc) → to be split into a
# separate always-fresh endpoint (Fase 5).
# - Backups appearing outside Monitor (cron/scheduled/retention)
# → client passes ?fresh=1 when its own cache is older than 6 h
# (Fase 4).
_VM_CACHE_INDEFINITE = 315_360_000 # 10 years — effectively infinite
_VM_DETAILS_TTL = _VM_CACHE_INDEFINITE
_VM_BACKUPS_TTL = _VM_CACHE_INDEFINITE
_VM_APPS_TTL = _VM_CACHE_INDEFINITE
_VM_SCHEDULE_TTL = _VM_CACHE_INDEFINITE
_VM_MOUNTS_TTL = _VM_CACHE_INDEFINITE
_vm_modal_cache_lock = threading.Lock()
def _vm_cache_get(cache: dict, vmid: int, ttl: int):
@@ -11506,7 +11525,14 @@ def _node_metrics_prewarmer_loop():
def _vm_modal_prewarmer_pass():
"""One full sweep of every guest's modal caches. Called both
from the startup warm-up and from the recurring loop. Returns
the number of guests successfully touched."""
the number of guests successfully touched.
The recurring loop is deliberately cheap: for each guest and
each cache we check the TTL FIRST and only invoke the handler
when the entry is actually stale. On steady state (all caches
fresh) a pass is O(guests) dict reads with no request contexts
created, no handlers entered, no pvesh spawned the CPU cost
disappears until something genuinely expires."""
from flask import g as _flask_g
warmed = 0
resources = get_cached_pvesh_cluster_resources_vm() or []
@@ -11515,83 +11541,74 @@ def _vm_modal_prewarmer_pass():
vm_type = r.get('type') # 'qemu' or 'lxc'
if vmid is None:
continue
try:
with app.test_request_context(f'/api/vms/{vmid}'):
_flask_g._internal_call = True
get_vm_config(vmid)
except Exception as e:
print(f"[ProxMenux] vm-modal prewarmer details {vmid}: {e}",
file=sys.stderr, flush=True)
try:
with app.test_request_context(f'/api/vms/{vmid}/backups'):
_flask_g._internal_call = True
api_vm_backups(vmid)
except Exception as e:
print(f"[ProxMenux] vm-modal prewarmer backups {vmid}: {e}",
file=sys.stderr, flush=True)
endpoints = [
(_vm_details_cache, _VM_DETAILS_TTL, get_vm_config,
f'/api/vms/{vmid}', 'details'),
(_vm_backups_cache, _VM_BACKUPS_TTL, api_vm_backups,
f'/api/vms/{vmid}/backups', 'backups'),
]
if vm_type == 'lxc':
endpoints.extend([
(_vm_apps_cache, _VM_APPS_TTL, api_vm_apps_get,
f'/api/vms/{vmid}/apps', 'apps'),
(_vm_schedule_cache, _VM_SCHEDULE_TTL, api_vm_apps_schedule,
f'/api/vms/{vmid}/schedule', 'schedule'),
(_vm_mounts_cache, _VM_MOUNTS_TTL, api_lxc_mount_points,
f'/api/lxc/{vmid}/mount-points', 'mounts'),
])
did_work = False
for cache, ttl, handler, route, label in endpoints:
if _vm_cache_get(cache, vmid, ttl) is not None:
continue # still fresh — skip the request-context overhead
try:
with app.test_request_context(f'/api/vms/{vmid}/apps'):
with app.test_request_context(route):
_flask_g._internal_call = True
api_vm_apps_get(vmid)
handler(vmid)
did_work = True
except Exception as e:
print(f"[ProxMenux] vm-modal prewarmer apps {vmid}: {e}",
file=sys.stderr, flush=True)
try:
with app.test_request_context(f'/api/vms/{vmid}/schedule'):
_flask_g._internal_call = True
api_vm_apps_schedule(vmid)
except Exception as e:
print(f"[ProxMenux] vm-modal prewarmer schedule {vmid}: {e}",
file=sys.stderr, flush=True)
try:
with app.test_request_context(f'/api/lxc/{vmid}/mount-points'):
_flask_g._internal_call = True
api_lxc_mount_points(vmid)
except Exception as e:
print(f"[ProxMenux] vm-modal prewarmer mounts {vmid}: {e}",
print(f"[ProxMenux] vm-modal prewarmer {label} {vmid}: {e}",
file=sys.stderr, flush=True)
warmed += 1
time.sleep(0.2) # brief breath so pvesh isn't hammered
if did_work:
time.sleep(0.2) # breath only when we actually ran a handler
return warmed
def _vm_modal_prewarmer_loop():
"""Keep the per-VM modal caches (details / backups / apps /
schedule) hot from the backend, so modals always open instantly
even after the browser tab has been closed for a long time.
The old React prefetcher only ran while the page was open.
"""One-shot warmup at service startup. Populates every per-VM
modal cache (details / backups / apps / schedule / mount points)
exactly once, then exits no periodic refresh loop.
Design (matches user's expectation of "heavy at startup, near-
zero during runtime"):
* One full warm-up pass right after startup so every cache is
primed before the user opens the UI.
* After that, a slow refresh loop at 180 s intervals. Since
the underlying TTLs are 5-15 min, the vast majority of
these calls are cache-hits (essentially free); real work
only happens when a cache is about to expire.
* Write actions (start/stop/reboot, apply update, edit
schedule) call `_vm_cache_invalidate(vmid, ...)` the
loop then refreshes just that guest on its next tick, and
an on-demand user open refreshes it immediately.
Rationale: every cache in this family is refreshed by explicit
event invalidation (see `_vm_cache_invalidate` calls scattered
across write endpoints). A periodic loop was double work and
lit up the CPU on hosts with many guests. The previous 5 min
tick meant ~500-1000 background subprocess/pvesh calls per hour
on a 25-guest host, entirely for data that hadn't changed.
LXC-only endpoints (apps, schedule) are skipped for qemu VMs."""
Trade-offs handled elsewhere:
* Backups added out-of-band (cron / scheduled / retention)
client passes `?fresh=1` on modal open when its local
cache is older than 6 h; server ignores the indefinite TTL
for that call and re-scans (Fase 4).
* Mount points runtime state (df/stat/ad-hoc) that changes
continuously served by a separate always-fresh endpoint
the client fetches on modal open (Fase 5).
Called once from the startup section. Thread exits after the
initial pass no `while True` loop."""
time.sleep(3) # let Flask finish binding before we invoke handlers
try:
t0 = time.time()
n = _vm_modal_prewarmer_pass()
print(f"[ProxMenux] VM-modal prewarmer: initial warm-up complete "
f"({n} guests in {time.time()-t0:.1f}s)", flush=True)
print(f"[ProxMenux] VM-modal prewarmer: warm-up complete "
f"({n} guests in {time.time()-t0:.1f}s) — no periodic refresh, "
f"caches held by event invalidation only", flush=True)
except Exception as e:
print(f"[ProxMenux] VM-modal prewarmer initial pass failed: {e}",
file=sys.stderr, flush=True)
while True:
time.sleep(180) # 3 min — most passes are cache-hits, near-zero cost
try:
_vm_modal_prewarmer_pass()
except Exception as e:
print(f"[ProxMenux] VM-modal prewarmer refresh error: {e}",
file=sys.stderr, flush=True)
@app.route('/api/node/metrics', methods=['GET'])
@@ -12569,11 +12586,26 @@ def api_create_backup(vmid):
@app.route('/api/vms/<int:vmid>/backups', methods=['GET'])
@require_auth
def api_vm_backups(vmid):
"""Get list of backups for a specific VM/LXC"""
"""Get list of backups for a specific VM/LXC.
The backend cache is indefinite (event-invalidated only). Out-of-
band backups cron jobs, scheduled vzdump, PBS retention pruning
never call `_vm_cache_invalidate`, so a naked GET would keep
serving the last snapshot for hours after a new file appeared.
To reconcile that without a background poll, the client tracks
the age of its own copy and, when older than its 6-hour gate,
calls this endpoint with `?fresh=1`. Server ignores the cached
entry for that call, re-scans every storage, writes the result
back into the cache and returns it. Subsequent openings within
the next 6 hours hit the freshened cache instantly.
"""
try:
cached = _vm_cache_get(_vm_backups_cache, vmid, _VM_BACKUPS_TTL)
if cached is not None:
return jsonify(cached)
force_fresh = request.args.get('fresh') in ('1', 'true', 'yes')
if not force_fresh:
cached = _vm_cache_get(_vm_backups_cache, vmid, _VM_BACKUPS_TTL)
if cached is not None:
return jsonify(cached)
backups = []
@@ -13928,15 +13960,17 @@ def get_vm_config(vmid):
@app.route('/api/lxc/<int:vmid>/mount-points', methods=['GET'])
@require_auth
def api_lxc_mount_points(vmid):
"""Sprint 13.29: per-LXC mount points enumeration.
"""Static half of the per-LXC mount-points payload — parsed mp
entries, source/target, PVE storage classification, host source
existence flags. Runtime state (`df` capacity, `stat` health,
ad-hoc NFS/CIFS discovery, runtime_mounted flag) lives in the
sibling `/api/lxc/<vmid>/mount-points/runtime` endpoint that
the client fetches on demand every time the tab opens.
Returns the parsed ``mpX:`` entries from the container config plus,
when the container is running, runtime status (mounted/not, real
fstype, options, stale detection) and any ad-hoc NFS/CIFS/SMB the
user mounted from inside the CT. Capacity is always populated from
the host-side source (PVE storage or `df` of the host path) so the
info is meaningful even on stopped containers.
"""
Backed by the indefinite `_vm_mounts_cache` (invalidated on
start/stop of the guest, since config-visible fields normally
only change through a guest reboot). The runtime endpoint is
NEVER cached it must reflect the live state at click time."""
cached = _vm_cache_get(_vm_mounts_cache, vmid, _VM_MOUNTS_TTL)
if cached is not None:
return jsonify(cached)
@@ -13945,7 +13979,7 @@ def api_lxc_mount_points(vmid):
except ImportError as e:
return jsonify({"ok": False, "error": f"helper unavailable: {e}"}), 503
try:
result = lxc_mount_points.get_lxc_mount_points(str(vmid))
result = lxc_mount_points.get_lxc_mount_points_static(str(vmid))
if not result.get("ok"):
return jsonify(result), 400
_vm_cache_put(_vm_mounts_cache, vmid, result)
@@ -13954,6 +13988,31 @@ def api_lxc_mount_points(vmid):
return jsonify({"ok": False, "error": str(e)}), 500
@app.route('/api/lxc/<int:vmid>/mount-points/runtime', methods=['GET'])
@require_auth
def api_lxc_mount_points_runtime(vmid):
"""Runtime half — always fresh, no cache. Returns per-target
runtime state + capacity, plus ad-hoc NFS/CIFS mounts detected
inside the running CT. Called by the client on every open of
the Mount Points tab so `df` usage and `stat` reachability are
real at click time; skips the whole prewarmer loop entirely.
Ad-hoc mounts and capacity are what the operator actually
watches (a stale NFS export shows here as `runtime_reachable
= false`), so caching them would defeat the point."""
try:
import lxc_mount_points
except ImportError as e:
return jsonify({"ok": False, "error": f"helper unavailable: {e}"}), 503
try:
result = lxc_mount_points.get_lxc_mount_points_runtime(str(vmid))
if not result.get("ok"):
return jsonify(result), 400
return jsonify(result)
except Exception as e:
return jsonify({"ok": False, "error": str(e)}), 500
@app.route('/api/vms/<int:vmid>/logs', methods=['GET'])
@require_auth
def api_vm_logs(vmid):
@@ -20595,7 +20654,7 @@ if __name__ == '__main__':
try:
vm_modal_thread = threading.Thread(target=_vm_modal_prewarmer_loop, daemon=True, name='vm-modal-prewarmer')
vm_modal_thread.start()
print("[ProxMenux] VM-modal prewarmer started (initial warm-up + 180s refresh)")
print("[ProxMenux] VM-modal prewarmer started (one-shot warm-up; caches refreshed by event invalidation only)")
except Exception as e:
print(f"[ProxMenux] VM-modal prewarmer failed to start: {e}")
+11 -1
View File
@@ -42,7 +42,17 @@ _APPS_DIR = "/etc/proxmenux/apps"
_PCT_BIN = "/usr/sbin/pct"
_PROBE_TIMEOUT_SEC = 15
_GITHUB_TIMEOUT_SEC = 15
_UPSTREAM_CACHE_TTL_SEC = 6 * 3600 # 6 h — GitHub is polite this way
# Aligned with the master LXC update cycle in
# notification_events.PollingCollector (UPDATE_CHECK_INTERVAL = 24 h).
# Previously this was 6 h — half a day out of sync with the apt/apk
# scan — so `refresh_all_apps` inside the 24 h collector would still
# hit GitHub for apps whose upstream TTL had elapsed, doubling
# checks. Unifying both to 24 h means one poll per day drives every
# update flavour (OS packages + community-scripts app upstream).
# Manual "Check" button + post-apply hook still pass force=True and
# ignore this TTL, so the user never has to wait for the timer to
# see a fresh result they explicitly asked for.
_UPSTREAM_CACHE_TTL_SEC = 24 * 3600
_VALID_METHODS = ("dpkg", "apk", "file", "binary",
"python_dist", "docker_label", "docker_exec",
+187 -8
View File
@@ -539,18 +539,197 @@ def _stat_via_host(host_pid: str, ct_target: str,
# ---------------------------------------------------------------------------
def get_lxc_mount_points(vmid: str) -> dict[str, Any]:
"""Top-level entry point used by the Flask route.
def get_lxc_mount_points_static(vmid: str) -> dict[str, Any]:
"""Static half of the mount-points payload — safe to cache
indefinitely because it only reads config and classifies against
PVE's storage inventory.
Returns:
- ``ok`` (bool)
- ``vmid`` (str)
- ``mount_points`` — list of configured mp0/mp1/... entries with
source / target / type / origin classification / host source
existence flags. No `df`, no `stat`, no ad-hoc detection.
The runtime enrichment (capacity, health, ad-hoc mounts,
runtime_mounted flag) lives in `get_lxc_mount_points_runtime`
and is fetched fresh on every modal open by the client. That
split lets the backend cache this half indefinitely (with event
invalidation on start/stop) while still giving the user real-
time capacity when they actually look."""
if not re.match(r"^\d+$", vmid):
return {"ok": False, "error": "invalid vmid"}
config_entries = _read_lxc_config(vmid)
pve_storages = _list_pve_storages()
out: list[dict[str, Any]] = []
for entry in config_entries:
source = entry.get("source", "")
target = entry.get("target", "")
cls = _classify(source, pve_storages)
host_src = _host_source_state(source)
out.append({
"mp_index": entry.get("mp_index", ""),
"source": source,
"target": target,
"type": cls["type"],
"origin_storage": cls.get("origin_storage", ""),
"origin_storage_type": cls.get("origin_storage_type", ""),
"origin_label": cls.get("origin_label", source),
"config_options": entry.get("config_options", {}),
"config_flags": entry.get("config_flags", []),
"host_source_exists": host_src["exists"],
"host_source_is_mountpoint": host_src["is_mountpoint"],
})
return {
"ok": True,
"vmid": vmid,
"mount_points": out,
}
def get_lxc_mount_points_runtime(vmid: str) -> dict[str, Any]:
"""Runtime half — always fresh, no cache. Fetched by the client
every time the Mount Points tab is opened so the operator sees
live capacity + reachability, plus any ad-hoc NFS/CIFS mounts
the container itself has made since the last static snapshot.
Returns:
- ``ok`` (bool)
- ``vmid`` (str)
- ``running`` (bool)
- ``mount_points`` — list of configured mp0/mp1/... entries
- ``ad_hoc`` — list of NFS/CIFS/SMB mounts found inside the running
CT that aren't backed by an mp config line
"""
# Validate vmid format — the value comes from a URL parameter, so
# we keep it strict to avoid path-traversal weirdness.
- ``runtime`` — dict keyed by target, containing runtime state
+ capacity per configured mount point
- ``ad_hoc`` — list of NFS/CIFS/SMB mounts done inside the CT
that aren't backed by an mp config line
The client merges `runtime[target]` onto the matching card from
the static payload; ad-hoc mounts render as their own cards
under a "Mounted inside container" divider. If the CT is down
or the client had no static payload for a target, the tab still
renders whatever runtime info is available (never blanks)."""
if not re.match(r"^\d+$", vmid):
return {"ok": False, "error": "invalid vmid"}
config_entries = _read_lxc_config(vmid)
pve_storages = _list_pve_storages()
running, host_pid = _ct_status(vmid)
rt_mounts = _read_ct_proc_mounts(host_pid) if running else []
# Same parallelisation as the pre-split path: `df`/`stat` per
# mount point are I/O-bound. Serialised, a CT with 5+ binds
# tripped Caddy's 3s reverse-proxy timeout.
from concurrent.futures import ThreadPoolExecutor
rt_by_target: dict[str, dict[str, Any]] = {m["rt_target"]: m for m in rt_mounts}
runtime_by_target: dict[str, dict[str, Any]] = {}
matched_targets: set[str] = set()
def _gather_one(entry):
src = entry.get("source", "")
tgt = entry.get("target", "")
classification = _classify(src, pve_storages)
capacity = _capacity_for(
src, classification, pve_storages,
config_options=entry.get("config_options", {}),
host_pid=host_pid if running else "",
target=tgt,
)
live_target = bool(running and tgt and tgt in rt_by_target)
health = _stat_via_host(host_pid, tgt) if live_target else None
return entry, capacity, live_target, health
if config_entries:
max_workers = max(2, min(8, len(config_entries)))
with ThreadPoolExecutor(max_workers=max_workers) as pool:
gathered = list(pool.map(_gather_one, config_entries))
else:
gathered = []
for entry, cap, live_target, health in gathered:
target = entry.get("target", "")
rt_item: dict[str, Any] = {**cap}
if live_target:
rt = rt_by_target[target]
rt_item.update({
"runtime_mounted": True,
"runtime_source": rt["rt_source"],
"runtime_fstype": rt["rt_fstype"],
"runtime_options": rt["rt_options"],
"runtime_readonly": rt["rt_readonly"],
"runtime_reachable": health["reachable"],
"runtime_error": health["error"],
})
matched_targets.add(target)
elif running:
rt_item["runtime_mounted"] = False
rt_item["runtime_error"] = "configured but not mounted"
else:
rt_item["runtime_mounted"] = None # CT down
runtime_by_target[target] = rt_item
# Ad-hoc remote mounts inside the running CT — same logic and
# parallelisation as before.
ad_hoc: list[dict[str, Any]] = []
if running:
ad_hoc_candidates = [
rt for rt in rt_mounts
if rt["rt_target"] not in matched_targets
and _REMOTE_FS_RE.match(rt["rt_fstype"])
]
if ad_hoc_candidates:
max_workers = max(2, min(8, len(ad_hoc_candidates)))
with ThreadPoolExecutor(max_workers=max_workers) as pool:
def _gather_adhoc(rt):
h = _stat_via_host(host_pid, rt["rt_target"])
if h.get("reachable"):
cap = _df_via_pct_exec(vmid, rt["rt_target"])
else:
cap = {"total_bytes": None, "used_bytes": None,
"available_bytes": None}
return rt, h, cap
results = list(pool.map(_gather_adhoc, ad_hoc_candidates))
for rt, health, cap in results:
ad_hoc.append({
"mp_index": "",
"source": rt["rt_source"],
"target": rt["rt_target"],
"type": "ad_hoc",
"origin_storage": "",
"origin_storage_type": "",
"origin_label": rt["rt_source"],
"config_options": {},
"config_flags": [],
"total_bytes": cap["total_bytes"],
"used_bytes": cap["used_bytes"],
"available_bytes": cap["available_bytes"],
"runtime_mounted": True,
"runtime_source": rt["rt_source"],
"runtime_fstype": rt["rt_fstype"],
"runtime_options": rt["rt_options"],
"runtime_readonly": rt["rt_readonly"],
"runtime_reachable": health["reachable"],
"runtime_error": health["error"],
})
return {
"ok": True,
"vmid": vmid,
"running": running,
"runtime": runtime_by_target,
"ad_hoc": ad_hoc,
}
def get_lxc_mount_points(vmid: str) -> dict[str, Any]:
"""Legacy combined entry point — kept for backwards compatibility
with any caller that still wants the pre-split shape. New code
should hit the static/runtime pair separately.
Merges the two halves so the returned dict matches what the
single-endpoint route used to return before the split."""
if not re.match(r"^\d+$", vmid):
return {"ok": False, "error": "invalid vmid"}