RedisState's compare-and-swap watches through its pipeline, so a write between the compare and EXEC fails the swap; pause_intake waits out a claim already under way, so nothing dispatches after it returns

This commit is contained in:
2026-11-04 21:53:41 +01:00
parent d028bf326a
commit 1a622edb1f
4 changed files with 74 additions and 32 deletions
+12 -3
View File
@@ -259,10 +259,19 @@ class ExecutionService:
"""Stop claiming, and wait for what is already claimed to finish.
Called around a rebuild: items claimed against the old pipeline should
finish there rather than half-run against the new one.
finish there rather than half-run against the new one. That includes
a claim still under way: the consumer holds its lock from the claim to
the dispatch, so once that lock is free here the claim is counted, and
every later pass finds the intake closed.
"""
self._intake.clear()
# So a consumer waiting for capacity sees the intake closed.
self._capacity_changed()
deadline = time.monotonic() + DRAIN_TIMEOUT_S
if not self._consumer_lock.acquire(timeout=DRAIN_TIMEOUT_S):
logger.warning("Rebuild did not wait out a claim")
return
self._consumer_lock.release()
with self._inflight_lock:
while self._running:
remaining = deadline - time.monotonic()
@@ -284,8 +293,8 @@ class ExecutionService:
"""Stop claiming, promoting, touching and reclaiming, once every
claimed cascade has finished. For a rename, which rewrites the queue.
Unlike `pause_intake` this also waits out a claim already blocking
when intake clears, and a timer pass. It waits without a limit and
Unlike `pause_intake` this also waits out a timer pass, and keeps the
consumer lock until `release`. It waits without a limit and
tells ``report`` the nodes whose cascades are still running each time
they change. Setting ``abort`` gives up: claiming resumes and it
returns False. True means held until `release`.