Postgres Postmaster does not scale
Summary
Recall.ai analyzes a PostgreSQL postmaster bottleneck under extreme scale, showing that the single-threaded postmaster loop and fork overhead can throttle new connections and parallel queries during bursts. They reproduce the issue with thousands of synchronized connections, profile the postmaster under load, and propose mitigations such as jitter in the fleet, avoiding bursts of parallel queries, plus kernel tuning like huge pages, concluding that postmaster contention is a key bottleneck in scaling PostgreSQL for real-time workloads.