[eumemic]

Automated dev-pipeline build for issue #2539.

[eumemic-bot]

Code review

• src/aios/db/procrastinate_extensions.py:77-99 — The catalog comparison is unlikely to ever recognize the installed objects as current because expected_functiondef() and expected_triggerdef() do not match PostgreSQL’s canonical pg_get_*def() output. In particular, pg_get_functiondef() includes the terminating statement semicolon after the dollar-quoted body, and pg_get_triggerdef() emits schema-qualified identifiers (including the executed function) in typical installations, while these helpers omit those forms. As a result, every aios migrate run will treat the trigger as drifted and execute the DROP TRIGGER/CREATE TRIGGER DDL, reacquiring an AccessExclusive lock and retaining the queue-stall regression this change is intended to prevent. Compare normalized/catalog-derived definitions or verify the exact server output rather than relying on these hand-constructed strings.

[eumemic]

Code review

Canonical review record (reconciler-written, company#383)

Verdict: MERGE-SAFE at b5d2943a4001ba09572d89a06b75bc7c782305b4.

No blocking findings at this head.

Canonical record written by the reconciler from the reviewer's structured return (company#383); the reviewer's own prose comment is discussion.

[eumemic]

Risk Assessment

Tier 4

Fix for #2539 (+441/-21 lines across 6 files). It changes how aios migrate handles the lock-release trigger on procrastinate_jobs, the busiest table in the system. The step now reads the Postgres catalog (pg_get_functiondef/pg_get_triggerdef) and skips the rewrite when the installed function and trigger already match. When they don't, it reapplies them with SET LOCAL lock_timeout and the same bounded retry Alembic uses (10 attempts, 1s apart); if those run out it raises, so the migration fails instead of stalling the job queue. Unit and integration tests are good: the unchanged-trigger case takes no lock, and a drifted trigger fails fast while a lock is held, then repairs. Rated high risk because this is migration/deploy code doing DDL that takes the strongest table lock (AccessExclusive) on the job queue table, and the skip depends on matching Postgres's exact text of the function and trigger, including schema and quoting. If that matching is wrong, migrate either silently skips the trigger update (lock releases would then only be noticed by the 5s polling backstop) or rewrites it every time. Migrate can also now fail on lock contention where it used to wait. Merge should be watched in a real deploy.

[eumemic]

Auto-merge receipt

The dev-pipeline reconciler auto-merged PR #2546 at head b5d2943a4001ba09572d89a06b75bc7c782305b4 after proving: risk tier 4 (cap 4), reviewed-green provenance for this head, the complete required check set green, no hold/fix/human veto, and behind_by=0 against master.

[eumemic]

Post-merge: a pipeline PR was merged WITHOUT the maker≠checker gate

pipeline:v2 PR #2546 MERGED without the reconciler's maker≠checker gate (risk:tier-4 is ABOVE the auto-merge cap (2)). Auto-merge requires reviewed:green + an in-cap risk tier — this merge bypassed it.