PREEMPT_NONE Is Dead; Your Postgres Probably Doesn’t Care
Summary
The article analyzes a PostgreSQL performance regression observed on Linux 7.0 due to the removal of PREEMPT_NONE, caused by minor page faults inside a spinlock when using large shared memory. It explains the mechanism, benchmark results, and practical mitigations (enable huge pages, kernel considerations, and container/vendor implications), concluding with actionable guidance for operators.