Pod in OOMKilled state. What to do?

If a Pod in Kubernetes is in the OOMKilled (Out Of Memory Killed) state, it means that the container exceeded the memory limit allocated to it in the specification and was terminated by the system.

What to do:

1. Check Pod logs and events:

   • kubectl describe pod <pod-name> — to see the reason for termination.
   • kubectl logs <pod-name> — to check what happened before OOM.

2. Analyze memory usage:

   • Use metrics (Prometheus, Metrics Server) to monitor memory consumption.

3. Configure memory limits and requests:

   • Ensure that resources.requests.memory and resources.limits.memory are set correctly.
   • Possibly increase the memory limit if the application requires more.

4. Optimize the application:

   • Check for memory leaks or inefficient usage.

5. Use profiling tools:

   • For Go — pprof, for Java — VisualVM, etc.

6. Consider horizontal scaling:

   • If the load is high, add replicas instead of increasing memory for a single Pod.

Thus, the main task is to understand the reason for exceeding memory and either optimize the application or adjust resource settings in Kubernetes.