fix(ingestion): R17 — structural unique index for one open job per document

[BigSimmo]

Summary

Implements R17 from docs/ingestion-concurrency-fix-workorder.md: a partial
unique index makes "at most one open ingestion job per document" a structural
DB invariant instead of relying on an application-side pre-check SELECT that
races the INSERT.

• supabase/migrations/20260708160000_ingestion_jobs_one_open_per_document.sql
  — create unique index concurrently if not exists ingestion_jobs_one_open_per_document_uidx on ingestion_jobs(document_id) where status in ('pending','processing').
• supabase/schema.sql — non-concurrent equivalent for fresh/scratch replay.
• src/app/api/documents/[id]/reindex/route.ts and
  src/app/api/documents/bulk/reindex/route.ts — on a 23505 unique violation
  from the job insert, both routes now return the same "already has pending or
  processing indexing work" 409 the pre-check produces, instead of a raw
  constraint 500.
• src/lib/ingestion-mutation-safety.ts — extracted buildActiveJobsSafetyResult()
  (shared by the pre-check and the new post-insert race handler) and exported
  activeIngestionJobColumns so both call sites select the same row shape.
• Tests added to tests/ingestion-mutation-safety.test.ts.

Known gap — do not merge yet

supabase/drift-manifest.json is not regenerated in this PR.
npm run drift:manifest requires Docker (it replays schema.sql into a
scratch container), and Docker is unavailable in the sandbox this PR was
authored in — Docker Desktop's WSL2 backend cannot start (Wsl/0x80070422),
confirmed unrecoverable after two restart attempts. tests/drift-detection.test.ts
compares the manifest's embedded schema.sql sha256 against the current file, so
CI's verify job will fail this PR as-is.

Next step: run npm run drift:manifest from an environment with a working
Docker daemon, commit the regenerated supabase/drift-manifest.json, then this
is ready for review/merge. Left as a draft PR for that reason.

Deploy order (per the work-order)

1. Merge this PR once drift-manifest is regenerated and CI is green (repo-only — no live DB write happens automatically).
2. Operator applies the migration's CONCURRENTLY statement manually against live (cannot run via supabase db push — CLI migrations are transactional and CONCURRENTLY cannot run inside one). Confirm npm run reindex:health reports jobs_pending = 0 and jobs_processing = 0 first (live had 0/0 as of 2026-07-08).
3. Run supabase migration repair --linked --status applied 20260708160000 so history matches effect.

Refs the phase-3 concurrency work-order (docs/ingestion-concurrency-fix-workorder.md).

🤖 Generated with Claude Code

[supabase]

Updates to Preview Branch (claude/r17-open-job-unique-index) ↗︎

Deployments	Status	Updated
Database	✅	Wed, 08 Jul 2026 16:34:07 UTC
Services	✅	Wed, 08 Jul 2026 16:34:07 UTC
APIs	✅	Wed, 08 Jul 2026 16:34:07 UTC

Tasks are run on every commit but only new migration files are pushed.
Close and reopen this PR if you want to apply changes from existing seed or migration files.

Tasks	Status	Updated
Configurations	✅	Wed, 08 Jul 2026 16:34:09 UTC
Migrations	❌	Wed, 08 Jul 2026 16:34:11 UTC
Seeding	⏸️	Wed, 08 Jul 2026 16:34:01 UTC
Edge Functions	⏸️	Wed, 08 Jul 2026 16:34:01 UTC

---

❌ Branch Error • Wed, 08 Jul 2026 16:34:12 UTC

ERROR: duplicate key value violates unique constraint "schema_migrations_pkey" (SQLSTATE 23505)
Key (version)=(20260708160000) already exists.
At statement: 3
INSERT INTO supabase_migrations.schema_migrations(version, name, statements) VALUES($1, $2, $3)

---

View logs for this Workflow Run ↗︎.
Learn more about Supabase for Git ↗︎.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.