fix(worker): reconcile ambiguous retry outcomes

[Joncallim]

Source Issue

Issue: #179

Parent

• Parent PR: fix(queue): upgrade legacy retry members atomically #295
• Base branch: codex/pr198-producer-5a2-r3a-retry-upgrade
• Parent/base SHA: 279801f5cf47e130b652724f437775d5d749dffb
• Child head SHA: 11d999123947f6f5465d1a754ba421324e8ef75a

Summary

When Redis may have applied retry-v3 but its response is lost, the worker replays the exact same retry transition once before closing the database attempt.

• If the original call or reconciliation replay succeeds, the attempt finishes as failed with the retained business error and Redis’s authoritative nextRetryAt.
• If both calls fail, the queue disposition remains unknown. The attempt now finishes honestly as terminal indeterminate, retaining the business error with no claimed retry deadline.
• If that best-effort attempt finalization also fails, the worker emits the existing fixed attempt-infrastructure diagnostic without exposing the caught error or job payload.

The dashboard presents indeterminate as Retry status unknown with warning styling. Task-attempt logs use warning-level queue.attempt.indeterminate evidence and the existing safe compatibility message. The final-attempt dead-letter path and queue/Lua implementation remain unchanged. PostgreSQL already stores attempt status as unconstrained text, so this type/documentation refinement requires no migration.

Local evidence

• Focused runtime and task-attempt suites: 25/25 passed
• Existing real-Redis occurrence/recovery proof: 3/3 passed on a disposable nonzero database
• Full unit command: 1,689 passed; only repository service-gated integrations were skipped locally
• Targeted ESLint: passed
• TypeScript --noEmit: passed
• Production build: passed
• git diff --check: passed
• Bare-return mutation: focused suite failed 2 tests as intended
• Ordinary failed plus null deadline mutation: focused suite failed 2 tests as intended

Hosted exact-head evidence

• Web run 30386306983, job 90366365188: passed
  • complete zero-skip unit suite: 114/114 files, 1,699/1,699 tests
  • mandatory S4 PostgreSQL: 14/14
  • mandatory S4 Redis scrub: 3/3
  • mandatory S4 Redis ACL: 3/3
  • production build: passed
  • mandatory S3 PostgreSQL: 16 passed
  • disabled-ingress proof: passed
  • fail-closed E2E bridge: 17 passed
• Contract run 30386325733, job 90366494118: passed
• GitGuardian Security Checks: passed

Scope

Only:

• web/worker/runtime.ts
• web/worker/task-attempts.ts
• web/db/schema.ts
• web/app/dashboard/tasks/[id]/page.tsx
• web/__tests__/core-diagnostic-output-closure.test.ts
• web/__tests__/task-attempts.test.ts

This PR remains draft and stacked on #295. Its unresolved review thread is intentionally left unanswered and unresolved for the Reviewer gate.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 554b97de4a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".