cybertec-postgresql / pg_timetable / 39578520812

88%

Build Type

push

github

Committed by web-flow

Commit Message

[-] fix `active_chain` row leak when a chain fails before its tasks run (#862)

* [-] fix `active_chain` row leak when chain cannot start or load tasks

`executeChain()` returned early on `StartTransaction()` and
`GetChainElements()` errors without calling `RemoveChainRunStatus()`, so the
row inserted by the worker stayed in `timetable.active_chain`. Once the
leaked rows reached `max_instances` the chain was never started again.

Remove the run status on both paths using a context detached from
cancellation, as the task failure path already does.

* [-] remove `active_chain` row in the worker so no `executeChain` exit can leak it

Both workers now call `removeChainRunStatus` right after `executeChain`
returns, pairing the delete with the insert in one place. Cleanup runs on
a detached context bounded by `log.DetachedTimeout` so a cancelled
scheduler context cannot skip it and a dead database cannot block a
worker forever.

Chains that fail before any task runs now count as failed: span status,
`RecordChainFailed` and the `on_error` handler fire on every failure path.
`StartTransaction` rolls back its own transaction when the vxid lookup
fails instead of handing a live `tx` back with an error.

* [-] start the cleanup timeout when cleanup starts, not when the chain starts

`executeChain` created its detached cleanup context before the chain's
transaction began, so `DetachedTimeout` was counted from the start of the
chain. Any chain running longer than 30s committed on an expired context:
pgx never sent `COMMIT`, the chain's transaction was rolled back, and the
chain was still logged as successful. Rollback, `on_error` and
self-destruct after 30s hit the same expired context.

Each exit path now creates its own detached context when its cleanup
starts. `DetachedTimeout` is a variable so the new test can shorten it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* [*] note why `DetachedTimeout` is a var

* [-] fix `RemoveChainRunStatus` droppi... (continued)

Coverage Stats

39 of 41 new or added lines in 5 files covered. (95.12%)

6 existing lines in 2 files now uncovered.

2265 of 2570 relevant lines covered (88.13%)

0.9 hits per line