Pods keep restarting with OOMKilled under load. How do you investigate and fix it?

Summary

Confirm with kubectl describe (reason OOMKilled, exit code 137), compare memory working set with the limit, and check how the runtime sizes its heap. Usually the heap is fixed above the container limit or non-heap memory wasn't accounted for. Size the runtime relative to the limit (MaxRAMPercentage, --max-old-space-size, GOMEMLIMIT), set requests and limits from measured usage, and look for leaks in GC metrics.

Longer Answer

OOM kills are SIGKILLs from the kernel: no graceful shutdown and no runtime exception. Restarts under load cascade as the remaining pods take more traffic, so add capacity first during an incident. Alert on restart reasons, not just availability.

Key Points

• 137 = SIGKILL, usually the memory limit.
• Runtime heap vs container limit.
• Size from measurements.
• Alert on OOMKilled restarts.

Deep dive: DevOps in Production: Restart Loop at Peak Traffic

What interviewers look for

A methodical investigation and container-aware runtime settings.