Question

A payments microservice running on GKE intermittently returns 5xx errors during peak traffic. In Cloud Monitoring, you notice the container's memory usage climbs steadily and pod restarts spike, while CPU utilization stays moderate. Kubernetes events show 'OOMKilled' for the affected pods. Your team needs to confirm the root cause with minimal code changes before deciding on a fix. What is the most effective next step?

Accepted Answer

Use Cloud Profiler to capture heap allocation profiles from the running service and correlate growing heap usage with the OOMKilled restart timestamps.

Answer

Increase the pod's CPU requests and limits, then redeploy and observe whether the 5xx errors stop.

Answer

Enable Cloud Trace sampling at 100% to identify which upstream request path is slow during peak traffic.

Answer

Create a log-based metric counting 5xx responses and set a burn-rate alert against the availability SLO.