Logical Replication Slot

A logical replication slot is a persistent structure on the PostgreSQL publisher that stores the WAL consumption position for each subscriber. It guarantees that no change is lost even if the subscriber disconnects temporarily: WAL segments are retained until they are consumed and acknowledged.

Why it exists #

Without a slot, PostgreSQL recycles WAL segments as soon as they’re no longer needed for crash recovery — typically within minutes on active systems. A subscriber disconnected for an hour would find an unrecoverable gap, and the only way out would be re-initialising the subscription from snapshot. The slot solves this by keeping the WAL available.

The orphaned slot risk #

A slot that stops consuming (crashed subscriber, dropped without removing the slot first, interrupted migration) keeps holding WAL indefinitely, filling up the publisher’s disk. It’s the number-one cause of outages from logical replication in production.

Essential monitoring #

The pg_replication_slots view exposes active (in use?), restart_lsn (where it would resume from), and computing the delta between pg_current_wal_lsn() and restart_lsn gives you the retained WAL volume. On critical systems an alert is mandatory when the delta exceeds a threshold (say 10 GB) or when a slot stays active = false for too long. PostgreSQL 13 introduced max_slot_wal_keep_size as a safety cap.