#829 completed_at metadata anomalies: claim-time timestamps on multi-minute runs; repair paths COALESCE stale values

closed medium engine Created 2026-07-19 05:26 · Updated 2026-07-19 06:06

Description

Edit
Observed: run 6d2f34ac showed completed_at ~1s after creation despite ~2min execution; reconciled zombies showed completed_at equal to claim time. Writers: orchestrator terminal writes (now), sync trigger COALESCE(completed_at,NOW()), zombie reconciler COALESCE, timeout handler COALESCE, jumper cancel_task COALESCE, first_start completed_at=NULL clear (now terminal-guarded + fail-open on a short conn — a skipped clear leaves stale values for every later COALESCE to preserve). Investigation + fix direction: repairs should stamp completed_at from the JUMPER run row's completed_at/failed_at/cancelled_at (authoritative), not COALESCE the possibly-stale workflow_run value; add an invariant check (non-terminal => completed_at IS NULL) to the reconciler.

Comments

Loading comments...

Context

Loading context...

Audit History

View All
Loading audit history...