Formal dead-letter queue + redrive. The one capability the docs concede is missing, and the most cross-confirmed in the market.
Evidence
- TaskQ's own docs: "no poison state and no dead-letter queue" (ops.md, deployment.md checklist).
- River Pro ships a dedicated dead-letter table designed for low index load (riverqueue.com/docs/pro/dead-letter-queue); pg-boss ships DLQ with redrive; Sidekiq OSS ships a Dead set with a Web UI tab, bulk restore, and long retention, and its changelog shows users demanded the bulk actions back.
Fix direction
Per-actor (or global) dead-letter routing target for jobs whose retry budget is exhausted; isolate to a dedicated low-index table (the jobs_archive machinery is the natural base); searchable list + bulk redrive in the admin UI; DLQ depth/age metrics added to the bundled alert rules. Mostly plumbing on existing primitives: the failure path, the archive table, and UI re-pending all exist.
Severity: high for adoption (closes the one self-declared no); low-medium effort.
Formal dead-letter queue + redrive. The one capability the docs concede is missing, and the most cross-confirmed in the market.
Evidence
Fix direction
Per-actor (or global) dead-letter routing target for jobs whose retry budget is exhausted; isolate to a dedicated low-index table (the jobs_archive machinery is the natural base); searchable list + bulk redrive in the admin UI; DLQ depth/age metrics added to the bundled alert rules. Mostly plumbing on existing primitives: the failure path, the archive table, and UI re-pending all exist.
Severity: high for adoption (closes the one self-declared no); low-medium effort.