feat: add pre-stop hook to unregister a compute node before its pod's termination

[arkbriar]

In addition,

• set the default terminationGracePeriodSeconds of compute pods to 10s to give the hook some time,
• bump the RisingWave to v1.8.0.

[arkbriar]

Tested locally.

[shanicky]

Unregistering workers will deregister the worker from meta. The corresponding worker will receive a notification that it has expired and then terminate itself. It is unclear whether early unregistering of the worker may cause the worker to quickly restart as a new one, rejoin, and then get killed by k8s 🤔

[arkbriar]

It is unclear whether early unregistering of the worker may cause the worker to quickly restart as a new one, rejoin, and then get killed by k8s 🤔

Actually it's well defined in the doc. It should be triggered when and only when k8s thinks the container needs to be terminated. It means the case could exist when a Pod fails to pass probe and gets restarted, and will quickly rejoin the cluster. IMO, it should be fine since the cluster will never be stable in any circumstances whenever a pod's killed. And also a cluster should converge toward a stable point when the pods are all running.

This hook is called immediately before a container is terminated due to an API request or management event such as a liveness/startup probe failure, preemption, resource contention and others.

https://kubernetes.io/docs/concepts/containers/container-lifecycle-hooks/#container-hooks

What I really concern about is the cost of recovery, as unregistration will definitely trigger one. There's no guarantee of how many times unregistration and recovery will happen before the it arrives the stable point.

[fuyufjh]

LGTM, but I believe this needs @shanicky's consent

[arkbriar]

Added an option computeComponent.autoDeregistration.enabled to control the behavior.

[shanicky]

LGTM