Replay dead-lettered records¶

Read what the writer set aside under dlq/ in the archive, fix the cause, and send the records through again.

Before you start¶

You need read access to the archive bucket and a writer to send to (--sink).
audit.writer.dead_lettered names the reason: unknown catalogue version, schema violation or oversized.
Replay one fixed cause at a time. A record that fails its schema is refused again, and should be.
An unknown catalogue version is dead-lettered and acknowledged, so no queue alarm sees it. Look for event=unknown_catalogue, and give the writer the version first. See change what a source records.
This is the archive's dlq/, not the SQS dead-letter queue. For that, see redrive the ingest DLQ.

audit replay --dlq --bucket <b> --from 2026-09-17 --to 2026-09-17

Fix the cause: register the catalogue version, configure the missing profile, or correct the emitter and deploy it.
Replay the one cause.

audit replay --dlq --bucket <b> --from 2026-09-17 --to 2026-09-17 \
  --reason "no catalogue" --sink https://audit-writer:8080

The command reports what came back under dlq/ and exits non-zero if anything did.

Run audit verify on the replayed range. Records keep their identifiers, so replaying one that got through is deduplicated.

None. A replay is idempotent. Report any record left under dlq/ to the emitter's owner with its reason: the writer will keep refusing it.