Exponential backoff is a retry strategy that increases the delay after each unsuccessful attempt. In Kubernetes, repeated container termination triggers progressively longer waits instead of immediate restarts, reducing rapid restart loops. section 6.2.1
Kubernetes applies the following restart sequence:
- Restart immediately after the first termination. section 6.2.1
- Wait 10 seconds, then double the delay after each subsequent termination: 20, 40, 80, and 160 seconds. section 6.2.1
- Cap the delay at 5 minutes; reset it to zero after the container runs successfully for 10 minutes. section 6.2.1
Observing a restart loop
After repeatedly sending Envoy's /quitquitquit endpoint, kubectl shows the pod as NotReady or CrashLoopBackOff while the container waits. The status message can report a value such as “back-off 40s restarting failed container=envoy.” section 6.2.1
$ kubectl get po kiada-ssl -o json | jq .status.containerStatuses
...
"state": {
"waiting": {
"message": "back-off 40s restarting failed container=envoy ...",
"reason": "CrashLoopBackOff"Inspect the container status to see the current backoff duration and waiting reason. section 6.2.1
