thermograph/singleton.py
Emi Griffith da1b2c8be1 Gate the subscription notifier to one worker (#161)
Prod runs 3 uvicorn workers, but the in-process subscription notifier was
started in every worker's lifespan (it was written assuming a single-worker
deploy). Its sweep fetches recent-forecast/archive data from Open-Meteo on a
15-min timer independent of any request, so running it 3× tripled the
background upstream load against a shared quota — the source of the overnight
429/503 rate-limit errors in logs/errors, and 3× redundant subscription scans.

Elect one worker to own the notifier via a non-blocking exclusive flock on a
lockfile (THERMOGRAPH_SINGLETON_LOCK): the first worker wins and holds the fd
for its lifetime; others stand down; the OS releases the lock if the leader
dies, so a restart re-elects cleanly. Unset (single worker / dev / tests)
always wins, so behavior there is unchanged. Mirrors the shared-metrics store's
env-selected cross-worker coordination.

The neighbor warmer stays per-worker on purpose: each worker drains its own
request-fed queue, so gating it would leave non-leader queues undrained.

- backend/singleton.py: flock-based leader election (claim()).
- backend/app.py: gate notify.start() on singleton.claim().
- deploy/thermograph.service: default THERMOGRAPH_SINGLETON_LOCK to
  data/notifier.lock (needs the unit reinstalled on prod to take effect).
- deploy/thermograph.env.example: document it alongside WORKERS.
- Tests: leader election — single-worker default, first-wins, idempotent
  re-claim, second-holder stands down, re-election after release. 183 passed.

Co-authored-by: root <root@vmi3417050.contaboserver.net>
2026-07-17 12:54:55 +00:00

50 lines
2.3 KiB
Python

"""Single-instance leader election for in-process background jobs.
With multiple uvicorn workers each worker process runs its own app lifespan, so a
background thread started there runs once *per worker*. That is correct for the
neighbor warmer (each worker drains its own request-fed queue) but wrong for the
subscription notifier: its periodic sweep scans every subscription and issues
upstream forecast/archive fetches on a timer, independent of any request. Running
it in N workers multiplies the Open-Meteo quota use N-fold (the source of the
overnight 429/503 rate-limit errors after prod went to 3 workers) and does the
same DB scan N times.
``claim`` elects one worker to own such jobs via a non-blocking exclusive advisory
lock (``flock``) on a lockfile:
* the first worker to call it acquires the lock and holds the fd for the process
lifetime, so the lock is never dropped while it lives -> returns ``True``;
* other workers fail the non-blocking lock and stand down -> return ``False``;
* the OS releases the lock when the leader exits, so a restart (or a crash) lets a
surviving/new worker win on its next call -> clean re-election, no stale state.
No lockfile configured (single-worker deploy, dev, tests) => always the leader.
This mirrors the metrics store's env-selected coordination: unset keeps the
zero-dependency single-worker behavior; a path opts into cross-worker arbitration.
"""
import fcntl
import os
# lock_path -> held fd. Kept open for the whole process lifetime: closing the fd (or
# letting it be GC'd) would drop the flock and let a second worker also become leader.
_held: dict[str, int] = {}
def claim(lock_path: "str | None") -> bool:
"""Return True if this process should own the single-instance job guarded by
``lock_path``. ``None``/empty (no multi-worker coordination configured) is always
the leader. Idempotent: re-claiming a path this process already holds returns True
without touching the lock."""
if not lock_path:
return True
if lock_path in _held:
return True
fd = os.open(lock_path, os.O_RDWR | os.O_CREAT, 0o644)
try:
fcntl.flock(fd, fcntl.LOCK_EX | fcntl.LOCK_NB)
except OSError:
# Another process already holds it — we're not the leader.
os.close(fd)
return False
_held[lock_path] = fd
return True