windmill

mirror of https://github.com/windmill-labs/windmill.git synced 2026-08-18 16:02:10 +00:00

Files

T

Ruben Fiszel 9762089fcb fix(schedules): re-arm enabled schedules left with no queued occurrence (#10174 )

* fix(schedules): re-arm enabled schedules left with no queued occurrence

* fix(schedules): lock schedule row while re-arming and report outcome

* fix(schedules): make reconcile lock cancellation-safe, re-check armed under lock

Address review feedback on the schedule reconciler:

- Use a transaction-scoped advisory lock (pg_try_advisory_xact_lock) instead of
  a session-scoped one. monitor_db runs under a 600s timeout; on cancellation a
  session lock on a pooled connection would be stranded, wedging reconciliation
  on every replica. An xact lock releases when its transaction is dropped.
- rearm_schedule re-checks for a queued occurrence under the row lock and
  returns NoOp if already armed, closing the scan→lock window that could
  double-push across a cron boundary. Add a regression test.
- Make reconcile_unarmed_schedules private (its only caller is in monitor.rs)
  and document its system-only contract.
- Log the disable only after the guarded UPDATE actually disables the schedule.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix(schedules): never disable from reconciliation and cap re-arms per pass

---------

Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

2026-07-17 18:08:10 +02:00

fixtures

feat: store hashed tokens instead of plaintext (#8217 )

2026-03-17 01:15:38 +00:00

concurrency_counter_test.rs

fix(concurrency): two-phase admit to skip FOR UPDATE on over-limit pulls (#9064 )

2026-05-07 09:43:33 +00:00

debounce_test.rs

fix(debounce): never supersede a running debounce survivor (#9780 )

2026-06-25 16:36:16 +00:00

job_cleanup_test.rs

refactor: move integration tests to subcrates to reduce recompilation