Skip to content

feat(reliability): dead-letter queue for poison events #14

@MarkCesium

Description

@MarkCesium

Summary

FastStream default behaviour on consumer exception is redelivery forever. A single malformed event halts progress on that partition.

Changes

  • Consumer wrapper: N retries with exponential backoff (default 5, 1s → 30s).
  • After N failures, publish the original message to devices.dlq with headers original_topic, error, traceback.
  • Prometheus counter devices_dlq_total{topic} and gauge devices_dlq_depth (see feat(observability): structured logging, metrics, tracing #19).

Verification

  • Feed a malformed message to places.member.added — after 5 retries it lands on devices.dlq, main topic progresses, metric increments. Alert from infra#19 fires.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions