All Your GUCs in a Row: log_lock_waits and log_lock_failures

log_lock_waits is PostgreSQL's cheapest lock-contention detector, riding the same timer that arms the deadlock detector (deadlock_timeout, one second by default). When on, it logs a 'still waiting' line with the blocking PID and queue, then an 'acquired' line with the total wait time when the lock finally comes through. It's off by default through version 18 but PostgreSQL 19 (currently beta 3) flips the default to on. A companion parameter, log_lock_failures, new in 18 and also off by default, logs who was holding a row when a SELECT ... FOR UPDATE NOWAIT fails outright rather than waiting. The piece walks through interactions with lock_timeout, deadlock_timeout's superuser-only context, and the narrow scope of log_lock_failures (row-level NOWAIT only, not LOCK TABLE or lock_timeout cancellations), recommending log_lock_waits=on everywhere and log_lock_failures=on except for NOWAIT-polling workloads.

Source: https://postgr.es/p/9tT. 8 Sync News only summarizes and links out; content copyright belongs to the authors and original sources.