Files
manage/backend/src/media_library_viewer_api/services/scheduler.py
T
2026-07-14 15:22:34 +00:00

297 lines
12 KiB
Python

"""Single-worker scheduler for typed backend actions."""
from __future__ import annotations
import logging
import threading
import time
from dataclasses import dataclass
from typing import Any
from media_library_viewer_api.observability import record_scheduled_action
from media_library_viewer_api.widgets.sources import ServiceRecord, build_service_record
from .scheduler_actions import ( # type: ignore[reportMissingImports]
QBITTORRENT_SPEED_ACTION,
get_scheduled_action,
)
from .scheduler_store import SchedulerRunStore # type: ignore[reportMissingImports]
from .settings_store import SettingsStore, get_settings_store
logger = logging.getLogger(__name__)
DEFAULT_INTERVAL_SECONDS = 15
MIN_INTERVAL_SECONDS = 5
MAX_INTERVAL_SECONDS = 300
DEFAULT_RETENTION_SECONDS = 1_800
MAX_RETENTION_SECONDS = 86_400
DEFAULT_MAX_ROWS = 1_200
MAX_MAX_ROWS = 1_200
SCHEDULER_LOOP_SECONDS = 1.0
BACKOFF_CAP_SECONDS = 300
class SchedulerBusyError(RuntimeError):
"""Raised when a manual action overlaps an existing service run."""
def _safe_int(value: Any, default: int = 0) -> int:
try:
return int(value)
except (TypeError, ValueError):
return default
def _bounded_config(config: dict[str, Any]) -> tuple[int, int, int]:
interval = max(
MIN_INTERVAL_SECONDS,
min(_safe_int(config.get("poll_interval_seconds"), DEFAULT_INTERVAL_SECONDS), MAX_INTERVAL_SECONDS),
)
retention = max(
60,
min(_safe_int(config.get("sample_retention_seconds"), DEFAULT_RETENTION_SECONDS), MAX_RETENTION_SECONDS),
)
max_rows = max(60, min(_safe_int(config.get("sample_max_rows"), DEFAULT_MAX_ROWS), MAX_MAX_ROWS))
return interval, retention, max_rows
@dataclass
class _ServiceState:
signature: tuple[Any, ...]
next_run_at: float | None = None
running: bool = False
last_attempt_at: int | None = None
last_success_at: int | None = None
last_error: str = ""
consecutive_failures: int = 0
backoff_until: int | None = None
class Scheduler:
"""Run registered service actions from one lifespan-managed worker."""
action_key = QBITTORRENT_SPEED_ACTION
def __init__(self) -> None:
self._thread: threading.Thread | None = None
self._stop_event = threading.Event()
self._state_lock = threading.RLock()
self._service_locks: dict[str, threading.Lock] = {}
self._states: dict[str, _ServiceState] = {}
self._run_store = SchedulerRunStore()
def start(self) -> None:
with self._state_lock:
if self._thread and self._thread.is_alive():
return
self._stop_event.clear()
self._thread = threading.Thread(target=self._run, name="scheduled-actions", daemon=True)
self._thread.start()
logger.info("Scheduled action worker started")
def stop(self, timeout: float = 5.0) -> None:
with self._state_lock:
thread = self._thread
if not thread:
return
self._stop_event.set()
thread.join(timeout=timeout)
if thread.is_alive():
logger.warning("Scheduled action worker did not stop within %.1fs", timeout)
else:
logger.info("Scheduled action worker stopped")
def status(self, service_id: str, store: SettingsStore | None = None) -> dict[str, Any]:
store = store or get_settings_store()
service = store.get_service(service_id)
if not service or service.get("service_type") != "qbittorrent":
raise ValueError("qBittorrent service not found")
config = service.get("config") or {}
interval, retention, max_rows = _bounded_config(config)
with self._state_lock:
state = self._states.get(service_id)
worker_running = bool(self._thread and self._thread.is_alive())
if state is None:
state = _ServiceState(signature=())
last_success = state.last_success_at
is_stale = last_success is None or time.time() - last_success > max(2 * interval, 60)
return {
"service_id": service_id,
"action_key": self.action_key,
"worker_running": worker_running,
"enabled": bool(service.get("enabled", True)) and bool(config.get("polling_enabled", True)),
"running": state.running,
"poll_interval_seconds": interval,
"sample_retention_seconds": retention,
"sample_max_rows": max_rows,
"next_run_at": _safe_int(state.next_run_at) if state.next_run_at is not None else None,
"last_attempt_at": state.last_attempt_at,
"last_success_at": last_success,
"last_error": state.last_error,
"consecutive_failures": state.consecutive_failures,
"backoff_until": state.backoff_until,
"is_stale": is_stale,
}
def run_now(self, service_id: str, store: SettingsStore | None = None) -> dict[str, Any]:
store = store or get_settings_store()
service_row = store.get_service(service_id)
if not service_row or service_row.get("service_type") != "qbittorrent":
raise ValueError("qBittorrent service not found")
if not service_row.get("enabled", True):
raise ValueError("Service is disabled")
config = service_row.get("config") or {}
if not config.get("polling_enabled", True):
raise ValueError("Polling is disabled")
service = build_service_record(store, service_row)
run = self._execute(service, "manual")
return {"run": run, "status": self.status(service_id, store)}
def _run(self) -> None:
while not self._stop_event.is_set():
try:
self._reconcile_and_run()
except Exception:
logger.exception("Scheduled action cycle failed")
self._stop_event.wait(SCHEDULER_LOOP_SECONDS)
def _reconcile_and_run(self) -> None:
store = get_settings_store()
services = sorted(store.list_services("qbittorrent"), key=lambda row: str(row.get("id") or ""))
active_ids = {str(row.get("id") or "") for row in services}
with self._state_lock:
for service_id in set(self._states) - active_ids:
self._states.pop(service_id, None)
self._service_locks.pop(service_id, None)
now = time.time()
for index, service_row in enumerate(services):
service_id = str(service_row.get("id") or "")
if not service_id:
continue
config = service_row.get("config") or {}
interval, _, _ = _bounded_config(config)
enabled = bool(service_row.get("enabled", True)) and bool(config.get("polling_enabled", True))
signature = (enabled, interval, config.get("sample_retention_seconds"), config.get("sample_max_rows"))
with self._state_lock:
state = self._states.get(service_id)
if state is None:
state = _ServiceState(signature=signature, next_run_at=now + min(index * 0.5, 5.0))
self._states[service_id] = state
elif state.signature != signature:
state.signature = signature
state.next_run_at = now if enabled else None
if not enabled:
state.next_run_at = None
continue
due = state.next_run_at is not None and now >= state.next_run_at
if due:
service = build_service_record(store, service_row)
try:
self._execute(service, "schedule")
except SchedulerBusyError:
logger.debug("Scheduled action already running service_id=%s", service_id)
with self._state_lock:
state = self._states.get(service_id)
if state:
delay = interval
if state.backoff_until:
delay = max(delay, state.backoff_until - _safe_int(time.time()))
state.next_run_at = time.time() + max(1, delay)
def _execute(self, service: ServiceRecord, trigger: str) -> dict[str, Any]:
lock = self._service_lock(service.id)
if not lock.acquire(blocking=False):
raise SchedulerBusyError(f"Action already running for service {service.id}")
try:
return self._execute_locked(service, trigger)
finally:
lock.release()
def _execute_locked(self, service: ServiceRecord, trigger: str) -> dict[str, Any]:
config = service.config
interval, _, _ = _bounded_config(config)
state = self._state_for(service.id, config)
attempt = state.consecutive_failures
start = time.perf_counter()
run = self._run_store.start_run(service.id, self.action_key, trigger, attempt=attempt)
with self._state_lock:
state.running = True
state.last_attempt_at = _safe_int(time.time())
action = get_scheduled_action(self.action_key)
try:
if action is None:
raise RuntimeError(f"Scheduled action is not registered: {self.action_key}")
action.run(service)
except Exception as exc:
duration_ms = _safe_int((time.perf_counter() - start) * 1000)
error = str(exc)[:1000]
finished = self._run_store.finish_run(run["id"], "failure", duration_ms=duration_ms, error=error)
self._run_store.prune(service.id, self.action_key)
with self._state_lock:
state.running = False
state.last_error = error
state.consecutive_failures += 1
delay = min(BACKOFF_CAP_SECONDS, max(interval, 2**state.consecutive_failures))
state.backoff_until = _safe_int(time.time()) + delay
record_scheduled_action(
service.id,
self.action_key,
"failure",
duration_seconds=duration_ms / 1000.0,
consecutive_failures=state.consecutive_failures,
)
logger.warning("Scheduled action failed service_id=%s action=%s: %s", service.id, self.action_key, error)
return finished if finished is not None else run
duration_ms = _safe_int((time.perf_counter() - start) * 1000)
finished = self._run_store.finish_run(run["id"], "success", duration_ms=duration_ms)
self._run_store.prune(service.id, self.action_key)
with self._state_lock:
state.running = False
state.last_success_at = _safe_int(time.time())
state.last_error = ""
state.consecutive_failures = 0
state.backoff_until = None
if trigger == "manual":
state.next_run_at = max(state.next_run_at or 0, time.time() + interval)
record_scheduled_action(
service.id,
self.action_key,
"success",
duration_seconds=duration_ms / 1000.0,
success=True,
consecutive_failures=0,
)
return finished if finished is not None else run
def _state_for(self, service_id: str, config: dict[str, Any]) -> _ServiceState:
with self._state_lock:
state = self._states.get(service_id)
if state is None:
interval, _, _ = _bounded_config(config)
state = _ServiceState(signature=(), next_run_at=time.time() + interval)
self._states[service_id] = state
return state
def _service_lock(self, service_id: str) -> threading.Lock:
with self._state_lock:
return self._service_locks.setdefault(service_id, threading.Lock())
_SCHEDULER = Scheduler()
def get_scheduler() -> Scheduler:
return _SCHEDULER
def reset_scheduler() -> None:
"""Reset the singleton for tests."""
global _SCHEDULER
_SCHEDULER.stop()
_SCHEDULER = Scheduler()
__all__ = ["Scheduler", "SchedulerBusyError", "get_scheduler", "reset_scheduler"]