Memory leak causes OOMKill loop

A service running on Kubernetes is in a restart loop. The pod dashboard shows containers being killed with reason 'OOMKilled' every 20–30 minutes, and memory usage climbs steadily from startup until it hits the container limit and the pod is killed and restarted. Request latency spikes each time a pod cycles. A new version shipped two days ago. CPU and request volume are normal. How do you triage and stop the churn?