fix(cleanup): compare live-paused statuses with IN, not ANY over a row constructor

[TheodoreSpeaks]

Summary

• unreferencedLargeValuePredicate() and the legacy/prune queries wrote pe.status = ANY(${LIVE_PAUSED_REFERENCE_STATUSES}::text[]). Drizzle expands a JS array to ($1, $2, $3) — correct for IN, but as ANY((...)::text[]) Postgres rejects it outright: cannot cast type record to text[]
• Switched all 9 sites to pe.status IN ${LIVE_PAUSED_REFERENCE_STATUSES}, which uses drizzle's native expansion as intended — no cast, no helper
• Broken since fix(large-refs): cleanup based on table read #4716 (2026-05-22), so cleanup-logs failed on every run since. The throw landed at step 2 of 6, which also silently disabled legacy large-value cleanup, large-value metadata pruning, job_execution_logs retention, and orphaned-snapshot cleanup

Type of Change

• Bug fix

Testing

• New live-paused-statuses-sql.test.ts renders the real unreferencedLargeValuePredicate() against the real drizzle dialect and schema (the global drizzle/schema mocks hide rendering bugs). Verified it fails on both the original ANY() form and on a status = IN (...) typo, and passes only on the correct form
• Executed the actual rendered predicate against staging Postgres: it runs and matches 52,203 unreferenced large values (the old form errors)
• 85 tests across lib/execution/payloads, background/cleanup-*, lib/cleanup pass; tsc --noEmit clean; bun run lint and the full audit suite pass

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Jul 28, 2026 4:41pm

[cursor]

PR Summary

Medium Risk
Touches retention/deletion eligibility for execution large values and metadata; wrong SQL would either fail jobs (old bug) or delete data that still references live paused runs.

Overview
Fixes broken cleanup-logs and large-value cleanup SQL by changing how LIVE_PAUSED_REFERENCE_STATUSES is compared in raw Drizzle sql fragments.

All nine predicates that gated deletion/pruning on “live” paused executions used status = ANY(${LIVE_PAUSED_REFERENCE_STATUSES}::text[]). Drizzle expands the JS array as a row constructor ($1, $2, $3), so Postgres rejected the statements (cannot cast type record to text[]) and the cleanup job failed early on every run.

Those sites now use status IN ${LIVE_PAUSED_REFERENCE_STATUSES}, which renders as a valid IN ($1, $2, $3) list. Updates span cleanup-logs.ts (legacy large-value file eligibility) and large-value-metadata.ts (stale reference prune + unreferencedLargeValuePredicate()).

Adds live-paused-statuses-sql.test.ts, which renders unreferencedLargeValuePredicate() with the real PgDialect (unmocked drizzle/schema) and asserts IN binding, no ANY(($… form, and no = IN typo.

^{Reviewed by Cursor Bugbot for commit f8df6e7. Bugbot is set up for automated code reviews on this repo. Configure here.}

[greptile-apps]

Greptile Summary

Fixes Postgres binding of live-paused statuses used in cleanup/prune SQL.

• Adds shared livePausedStatusArray that renders ARRAY[$n,...]::text[] via drizzle sql.join instead of a row constructor from a raw JS array.
• Replaces all nine = ANY(...) sites in cleanup-logs and large-value metadata helpers; leaves notInArray(..., LIVE_PAUSED_REFERENCE_STATUSES) unchanged.
• Adds dialect-level render tests and mocks livePausedStatusArray in cleanup-logs tests.

Confidence Score: 5/5

Safe to merge; the SQL array-binding fix matches the failure mode and all prior broken ANY() sites are covered.

Shared fragment produces ARRAY[...]::text[] with one bind per status; remaining LIVE_PAUSED_REFERENCE_STATUSES use is notInArray, which expects a JS array; tests pin the render shape against the real dialect.

Important Files Changed

Filename	Overview
apps/sim/lib/execution/payloads/large-value-metadata.ts	Introduces correctly rendered text[] fragment and updates all ANY() guards that previously cast a row constructor.
apps/sim/background/cleanup-logs.ts	Switches four ANY() predicates to livePausedStatusArray; notInArray still uses the JS status list appropriately.
apps/sim/lib/execution/payloads/live-paused-status-array.test.ts	Unmocks drizzle and asserts ARRAY rendering, param count, and rejection of the broken row-constructor form.
apps/sim/background/cleanup-logs.test.ts	Mocks the new livePausedStatusArray export so cleanup-logs unit tests keep loading.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[LIVE_PAUSED_REFERENCE_STATUSES JS array] --> B[livePausedStatusArray sql fragment]
  B --> C["ANY(ARRAY[$1,$2,$3]::text[])"]
  C --> D[cleanup-logs legacy large values]
  C --> E[unreferencedLargeValuePredicate]
  C --> F[pruneStaleReferences]

Loading

_{Reviews (1): Last reviewed commit: "fix(cleanup): bind live-paused statuses ..." | Re-trigger Greptile}