File descriptor exhaustion is a common but misdiagnosed cause of PostgreSQL outages. The kernel enforces both a system-wide FD ceiling (fs.file-max) and a per-process limit (RLIMIT_NOFILE). PostgreSQL's process-per-connection model means high connection counts — especially without a pooler — can exhaust these limits fast, particularly when lock contention extends how long backends hold FDs open. The post walks through the exact log sequence during an outage, OS and PostgreSQL-level diagnostic commands, and the proper fix: deploying PgBouncer to multiplex thousands of app connections down to a small backend pool. Raising fs.file-max is covered as a temporary measure, including how to apply it on Kubernetes (AKS) via kubectl debug or a privileged DaemonSet. Monitoring recommendations focus on alerting before file-nr approaches fs.file-max rather than reacting after errors appear.
