Slot Machine

A Debezium Postgres connector has been stopped for a two-day maintenance window. Write volume is normal, yet pg_wal on the primary keeps growing and the instance is close to running out of disk. What is the cause?

The connector's initial snapshot writes its own copy of every scanned row back into the WAL, and that snapshot never completed before the maintenance window began

The connector's inactive logical replication slot still pins its restart position, so Postgres cannot recycle any WAL segment written since the connector stopped

Setting wal_level to logical roughly doubles the volume of WAL written per transaction, so two ordinary days of traffic produced more than the disk could hold

A long-running transaction is holding back the vacuum horizon, which forces Postgres to keep every WAL segment written since that transaction began