Failover and Recovery¶

Postgres Stream handles both sink failures and slot invalidation automatically.

How It Works¶

Events are stored in a partitioned table, not just the WAL. This means:

• WAL can be released immediately
• Recovery reads from the table, not the WAL
• 7-day partition retention = 7-day recovery window

Sink Failure¶

When the sink is unavailable:

1. Checkpoint saved at failed event
2. Retries until sink recovers
3. Replays missed events via COPY
4. Resumes normal streaming

Slot Invalidation¶

When the replication slot is invalidated:

1. Queries last confirmed LSN from invalidated slot
2. Sets checkpoint to first event after that LSN
3. Drops slot and restarts pipeline
4. Replays from checkpoint

No operator intervention required.

Guarantees¶

• Zero data loss - As long as downtime < partition retention (7 days)
• At-least-once delivery - Events may repeat during replay
• Order preserved - Events replay in order