Container Size Calculator

Use kubectl top pods, metrics-server, or Prometheus. Look at P95 actual usage over a week. Set requests to P95 usage + 20% buffer.

The JVM has heap memory, metaspace, thread stacks, GC overhead, and native memory. A "256MB" Java app often needs 512MB-1GB total. Set -Xmx to 75% of memory limit.

1:2 for production (safe). 1:4 for development (efficient). 1:1 for critical services (Guaranteed QoS). Never set requests higher than limits.

Set memory limits to catch leaks early (container gets OOMKilled and restarted). Use readiness probes so traffic shifts before the restart.

VPA automatically adjusts requests based on actual usage. Good for right-sizing but can cause pod restarts. Use in "recommend" mode first.

Container exceeded its memory limit. Increase memory limit, fix memory leaks, or reduce in-process caching. Check: kubectl describe pod for OOMKilled status.

Insufficient cluster resources to satisfy requests. Either reduce requests, add nodes, or check resource quotas in the namespace.

CPU throttling. Kubernetes limits CPU in 100ms windows. Your app may spike to limit in bursts. Increase CPU limit or remove it.

OOMKilled (memory), failed liveness probe (too slow to respond), or application crash. Check kubectl describe pod and kubectl logs.

Warden monitors your services externally. When pods are OOMKilled or throttled, Warden detects the degraded response times and alerts you before users complain.