Resolve OOMKilled Errors by Reading VPA Memory Recommendations
Scenario
A Deployment web-app is running in the production namespace with 2 replicas.
The web-container consumes more memory than the configured limit, causing repeated OOMKilled events.
A Vertical Pod Autoscaler (VPA) web-app-vpa is already configured for this Deployment with updateMode: Off, meaning it only provides resource recommendations and does not automatically update the Deployment.
Task
- Check the VPA recommendation for the
web-container. - Update the Deployment
web-appto apply the VPA Target recommended memory request and memory limit. - Ensure the Deployment rollout completes successfully and the replica count remains
2.
Do not modify CPU values (they must remain 50m).
Wait a few minutes for the VPA to collect metrics before checking recommendations.
Hint 1
Use kubectl describe pod to identify OOMKilled containers and kubectl describe vpa web-app-vpa -n production to view resource recommendations. Apply the recommended values using kubectl set resources or by editing the Deployment directly.
See the official docs: Vertical Pod Autoscaler