PG-SLOTS-001 · Replication slots retaining WAL, or replicas lagging

Severity	Confidence	Permission	Applies to
High	Hint · Confirmed above 10 GB retained or wal_status lost/unreserved	pg_monitor	PostgreSQL 14+, RDS, Aurora, Azure Flexible (on the primary)

What it checks

• pg_replication_slots: inactive slots retaining at least 1 GB of WAL (retained_mb), or any slot whose wal_status is unreserved or lost.
• pg_stat_replication: replicas whose replay_lag is at least 60 s (lag_s).

Why it matters

A slot keeps every WAL segment its consumer hasn't confirmed. If the consumer is gone (an old replica, a CDC tool that was turned off), WAL piles up until the disk fills and the primary stops accepting writes.

Next action

-- slot no longer needed
SELECT pg_drop_replication_slot('<slot_name>');

For lagging replicas, check the replica's disk and CPU, the network, and long queries on the replica that delay replay (max_standby_streaming_delay). Consider max_slot_wal_keep_size as a safety limit.

Limitations

Only meaningful on the primary. A slot may be inactive for a few seconds while its consumer reconnects.