When a Pod is deleted, Kubelet (the agent running on each AKS node) handles its graceful termination using a process called graceful shutdown. Here’s how it works:
1. User runs kubectl delete pod
You (or a controller like Deployment) run:
kubectl delete pod my-app
This sends a delete request to the Kubernetes API server.
2. API server sets a “deletion timestamp”
The pod is not immediately removed. Instead, Kubernetes marks it for deletion with a terminationGracePeriodSeconds (default: 30 seconds).
This is a signal to Kubelet to begin shutting it down gracefully.
3. Kubelet receives the deletion event
Kubelet notices the pod has a deletion timestamp and begins termination.
💬 4. Kubelet sends SIGTERM to containers
Kubelet sends a SIGTERM (terminate signal) to each container in the pod.
- Containers should catch this signal and begin shutting down.
- For example, a web server may finish processing current requests.
5. Wait for terminationGracePeriodSeconds
Kubelet gives the container time to exit gracefully (default 30s, can be customized in Pod spec):
spec:
terminationGracePeriodSeconds: 10
6. If container doesn’t stop → send SIGKILL
If the container doesn’t exit in time, Kubelet forcefully stops it using SIGKILL.
This is like a hard shutdown (kill -9).
✅ 7. Kubelet deletes the Pod from the node
Once containers are stopped (gracefully or forcefully), Kubelet cleans up:
- Removes the pod’s containers and volumes
- Frees node resources
- Updates the status back to the API server
👀 Optional: preStop Hook
You can run a custom script before shutdown using preStop:
lifecycle:
preStop:
exec:
command: [“/bin/sh”, “-c”, “echo Shutting down… && sleep 5”]
This runs before SIGTERM is sent.
