- Check
GET /healthand inspect the latest row inpipeline_runs. - Run
c360 process; pending Bronze events are safe to replay. - Run
c360 validateto check identity, SCD, and profile invariants. - For a disposable local reset, run
make seed. Do not use this reset on persistent environments.
- Record connector/task state and
MilliSecondsBehindSourcebefore restarting. - Check PostgreSQL replication-slot retained bytes or MySQL binlog availability.
- Restart only the failed task. Do not delete a slot or connector offset unless a controlled resnapshot has been approved.
- Confirm offsets advance, DLQ remains empty, and source/Bronze reconciliation returns to zero.
- Stop publishing the affected xref run to Gold.
- Inspect
identity_quarantineand the high-degree identifier. - Add a reviewed block/guard rule; never manually edit only the Gold profile.
- Replay identity edges, compare precision/recall, publish remaps, then rebuild affected profiles.
Pin the input Bronze snapshot/offset range and identity-model version. Write to an Iceberg branch, run reconciliation and no-future-leakage checks, then fast-forward after approval. Backfills must not overwrite raw Bronze data.
Resolve all identifiers to the master ID, tombstone source records, purge Redis keys, delete affected Iceberg rows, expire snapshots after the legal retention hold, and write an immutable audit event containing no erased PII.