Parameter · Vacuuming / Cost-Based Vacuum Delay

vacuum_cost_limit

Vacuum cost amount available before napping.

Default: 200
Type: integer
Range: 1 … 10000
Applies: Session (any user)

In plain words

The size of VACUUM's budget between naps. A bigger budget means more pages per nap, so a faster but more I/O-hungry vacuum. Autovacuum uses this value too unless autovacuum_vacuum_cost_limit is set.

For DBAs and operators

Because autovacuum_vacuum_cost_limit defaults to -1, this 200 is effectively the total budget shared by all autovacuum workers. With the 2 ms autovacuum delay, 200 points allows only about 40 MB/s of dirtied pages for all workers together, often too little for large, busy databases on SSDs.

Typical tuning

Modern SSD/NVMe: 1000–4000 (set autovacuum_vacuum_cost_limit explicitly so manual VACUUM is unaffected).
Watch disk utilisation and replica lag after raising it: vacuum also generates WAL.
Signs it is too low: autovacuum workers always busy, wait_event = VacuumDelay in pg_stat_activity, rising n_dead_tup.

SELECT pid, wait_event, query FROM pg_stat_activity
 WHERE backend_type = 'autovacuum worker';

Since PostgreSQL 16 a config reload updates the cost settings of already-running autovacuum workers.

How to change it

Any user can change it for their session with SET.

See it in context

VACUUM Autovacuum: when tables get vacuumed

Related parameters

autovacuum_vacuum_cost_limit vacuum_cost_delay vacuum_cost_page_dirty autovacuum_max_workers

All Vacuuming parameters