hard20 pts · first answer countspostgresqlperformance

OOM killer after raising work_mem

To speed up reports you set work_mem = 1GB globally on a server with 64 GB RAM and 200 connections. A week later the OOM killer stops PostgreSQL. Why?

A.work_mem is shared by all sessions, so 1 GB cannot cause OOMB.work_mem only affects VACUUMC.Each sort or hash operation in each session may use up to work_mem (hash more with hash_mem_multiplier), so many concurrent complex queries can use far more than RAMD.The OOM killer only runs when shared_buffers is too small

← Rejoining the old primary Restore takes longer than the RTO →