Our Production Kubernetes Pods Kept Restarting | OOMKilled #tech #software #programming

If you think Kubernetes kills pods randomly, you're calculating memory wrong. 📉

We kept seeing OOMKilled errors even though our heap size was within limits. The mistake? We treated the heap as the only memory consumer.

The Reality of Container Memory:

Your limit needs to cover the entire process, not just the heap:

Metaspace: Class metadata isn't free.

Thread Stacks: Each thread eats stack memory.

Direct Buffers: NIO uses off-heap memory.

When you set limits: 2Gi and Xmx2g, you are guaranteeing a crash.

The kernel doesn't care about your heap; it cares about the contract.

Fix: Measure real usage (RSS), not just heap, and leave room for the spikes.
#kubernetes #devops #softwareengineer

Видео Our Production Kubernetes Pods Kept Restarting | OOMKilled #tech #software #programming канала Mohinish Joshi