Files
wehub-resource-sync 6ede33ccdb
Build and Push Docker Images / create_manifest (web, surfsense-web, , cpu) (push) Has been cancelled
Build and Push Docker Images / finalize_release (push) Has been cancelled
Obsidian Plugin Lint / lint (push) Has been cancelled
Build and Push Docker Images / build (./surfsense_web, cpu, ./surfsense_web/Dockerfile, web, surfsense-web, ubuntu-24.04-arm, linux/arm64, arm64, , runner, false, cpu) (push) Has been cancelled
Build and Push Docker Images / build (./surfsense_web, cpu, ./surfsense_web/Dockerfile, web, surfsense-web, ubuntu-latest, linux/amd64, amd64, , runner, false, cpu) (push) Has been cancelled
Build and Push Docker Images / compute_version (push) Has been cancelled
Build and Push Docker Images / build (./surfsense_backend, cpu, ./surfsense_backend/Dockerfile, backend, surfsense-backend, ubuntu-24.04-arm, linux/arm64, arm64, , production, false, cpu) (push) Has been cancelled
Build and Push Docker Images / build (./surfsense_backend, cpu, ./surfsense_backend/Dockerfile, backend, surfsense-backend, ubuntu-latest, linux/amd64, amd64, , production, false, cpu) (push) Has been cancelled
Build and Push Docker Images / build (./surfsense_backend, cu126, ./surfsense_backend/Dockerfile, backend, surfsense-backend, ubuntu-24.04-arm, linux/arm64, arm64, -cuda126, production, true, cuda126) (push) Has been cancelled
Build and Push Docker Images / build (./surfsense_backend, cu126, ./surfsense_backend/Dockerfile, backend, surfsense-backend, ubuntu-latest, linux/amd64, amd64, -cuda126, production, true, cuda126) (push) Has been cancelled
Build and Push Docker Images / build (./surfsense_backend, cu128, ./surfsense_backend/Dockerfile, backend, surfsense-backend, ubuntu-24.04-arm, linux/arm64, arm64, -cuda, production, true, cuda) (push) Has been cancelled
Build and Push Docker Images / build (./surfsense_backend, cu128, ./surfsense_backend/Dockerfile, backend, surfsense-backend, ubuntu-latest, linux/amd64, amd64, -cuda, production, true, cuda) (push) Has been cancelled
Build and Push Docker Images / verify_digests (push) Has been cancelled
Build and Push Docker Images / create_manifest (backend, surfsense-backend, , cpu) (push) Has been cancelled
Build and Push Docker Images / create_manifest (backend, surfsense-backend, -cuda, cuda) (push) Has been cancelled
Build and Push Docker Images / create_manifest (backend, surfsense-backend, -cuda126, cuda126) (push) Has been cancelled
chore: import upstream snapshot with attribution
2026-07-13 13:33:44 +08:00

165 lines
6.1 KiB
Python

"""Celery tasks package.
Also hosts the small helpers every async celery task should use to
spin up its event loop. See :func:`run_async_celery_task` for the
canonical pattern.
"""
from __future__ import annotations
import asyncio
import contextlib
import logging
from collections.abc import Awaitable, Callable
from typing import TypeVar
from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
from sqlalchemy.pool import NullPool
from app.config import config
logger = logging.getLogger(__name__)
_celery_engine = None
_celery_session_maker = None
def get_celery_session_maker() -> async_sessionmaker:
"""Return a shared async session maker for Celery tasks.
A single NullPool engine is created per worker process and reused
across all task invocations to avoid leaking engine objects.
"""
global _celery_engine, _celery_session_maker
if _celery_session_maker is None:
# Reap connections orphaned mid-transaction (e.g. a worker that hung or
# crashed mid-index) so they can't hold locks on documents/chunks and
# wedge writes — the failure mode that previously left an "idle in
# transaction" session holding locks for 11+ hours. Kept generous so a
# legitimate long per-document embed window is never killed.
connect_args: dict = {}
idle_ms = config.DB_CELERY_IDLE_IN_TX_TIMEOUT_MS
if (
idle_ms
and idle_ms > 0
and config.DATABASE_URL
and "asyncpg" in config.DATABASE_URL
):
connect_args["server_settings"] = {
"idle_in_transaction_session_timeout": str(idle_ms)
}
_celery_engine = create_async_engine(
config.DATABASE_URL,
poolclass=NullPool,
echo=False,
connect_args=connect_args,
)
with contextlib.suppress(Exception):
from app.observability.bootstrap import instrument_sqlalchemy_engine
instrument_sqlalchemy_engine(_celery_engine)
_celery_session_maker = async_sessionmaker(
_celery_engine, expire_on_commit=False
)
return _celery_session_maker
def _dispose_shared_db_engine(loop: asyncio.AbstractEventLoop) -> None:
"""Drop the shared ``app.db.engine`` connection pool synchronously.
The shared engine (used by ``shielded_async_session`` and most
routes / services) is a module-level singleton with a real pool.
Each celery task creates a fresh ``asyncio`` event loop; asyncpg
connections cache a reference to whichever loop opened them. When
a subsequent task's loop pulls a stale connection from the pool,
SQLAlchemy's ``pool_pre_ping`` checkout crashes with::
AttributeError: 'NoneType' object has no attribute 'send'
File ".../asyncio/proactor_events.py", line 402, in _loop_writing
self._write_fut = self._loop._proactor.send(self._sock, data)
or hangs forever inside the asyncpg ``Connection._cancel`` cleanup
coroutine that can never run because its loop is gone.
Disposing the engine forces the pool to drop every cached
connection so the next checkout opens a fresh one on the current
loop. Safe to call from a task's finally block; failure is logged
but never propagated.
"""
try:
from app.db import engine as shared_engine
loop.run_until_complete(shared_engine.dispose())
except Exception:
logger.warning("Shared DB engine dispose() failed", exc_info=True)
def _dispose_shared_checkpointer_pool(loop: asyncio.AbstractEventLoop) -> None:
"""Drop the shared checkpointer pool so the next task opens a fresh one.
The durable ``AsyncPostgresSaver`` pool binds asyncpg connections to the
loop that opened them; a later task on a fresh loop stalls on a stale
connection (``PoolTimeout``). Failure is logged, not raised.
"""
try:
from app.agents.chat.runtime.checkpointer import close_checkpointer
loop.run_until_complete(close_checkpointer())
except Exception:
logger.warning("Shared checkpointer pool dispose() failed", exc_info=True)
T = TypeVar("T")
def run_async_celery_task[T](coro_factory: Callable[[], Awaitable[T]]) -> T:
"""Run an async coroutine inside a fresh event loop with proper
DB-engine cleanup.
This is the canonical entry point for every async celery task.
It performs three responsibilities that were previously copy-pasted
(incorrectly) across each task module:
1. Create a fresh ``asyncio`` loop and install it on the current
thread (celery's ``--pool=solo`` runs every task on the main
thread, but other pool types don't).
2. Dispose the shared ``app.db.engine`` BEFORE the task runs so
any stale connections left over from a previous task's loop
are dropped — defends against tasks that crashed without
cleaning up.
3. Dispose the shared engine AFTER the task runs so the
connections we opened on this loop are released before the
loop closes (avoids ``coroutine 'Connection._cancel' was
never awaited`` warnings and the next-task hang).
Use as::
@celery_app.task(name="my_task", bind=True)
def my_task(self, *args):
return run_async_celery_task(lambda: _my_task_impl(*args))
"""
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
try:
# Defense-in-depth: prior task may have crashed before
# disposing. Idempotent — no-op if pool is already empty.
_dispose_shared_db_engine(loop)
_dispose_shared_checkpointer_pool(loop)
return loop.run_until_complete(coro_factory())
finally:
# Drop any connections this task opened so they don't leak
# into the next task's loop.
_dispose_shared_db_engine(loop)
_dispose_shared_checkpointer_pool(loop)
with contextlib.suppress(Exception):
loop.run_until_complete(loop.shutdown_asyncgens())
with contextlib.suppress(Exception):
asyncio.set_event_loop(None)
loop.close()
__all__ = [
"get_celery_session_maker",
"run_async_celery_task",
]