failure-reporter

Cloudflare Worker that consumes the aauth-events-dlq dead-letter queue and emits a metadata-only aauth.shipping_failed event back into the main aauth-events queue.

Why

Events that fail to ship (the shipper Worker exhausts ~1 hour of retries against Freezer) land in aauth-events-dlq and would otherwise vanish silently from queryable Parquet. This Worker synthesizes a minimal failure event for each one — original_event_id, original_event, original_timestamp, original_service — so the gap is itself a row you can SELECT against.

We deliberately do not re-emit the original payload: poison events would hit the DLQ again on the same failure. The DLQ itself retains full bodies for 4 days for dashboard inspection if you need to see what was lost.

Deploy

npm install
npx wrangler deploy

No secrets needed. Cloudflare wires the consumer (aauth-events-dlq) and producer (aauth-events) bindings from wrangler.toml at deploy time.

Verify

When events land in the DLQ (e.g. Freezer is down for >1 hour), they'll be reprocessed and a aauth.shipping_failed event will appear in Parquet. Query example:

SELECT
  timestamp,
  json_extract_string(data, '$.original_event') AS original_event,
  json_extract_string(data, '$.original_event_id') AS original_event_id
FROM events
WHERE event = 'aauth.shipping_failed'
ORDER BY timestamp DESC
LIMIT 20;

See AAuth-dev/EVENT-LOGGING-PLAN.md for the broader design.