When scale replicas==0, do K8s send SIGTERM?

General Discussions
1

Hello guys,

I didn’t find this information at K8s doc:

When we are deleting a pod, we have all the lifecycle of delete (one of it’s step is to send SIGTERM to the pod application).

But when we are “faking” the restart, using:

kubectl scale --replicas=0 ...
kubectl scale --replicas=1 ...

Are we receiving the same SIGTERM?

I’m asking because my application doesn’t seems to receive the signal when we are scalling…

2

Yes, there’s not really a way to bypass that

3

It should receive SIGTERM and if its taking too long, it will eventually get a SIGKILL. You can adjust the timeout with terminationGracePeriodSeconds.

For a detailed explanation of the process and how to modify it, checkout the Termination of Pods section of the docs.

4

This SIGTERM and SIGKILL is when we are talking about:

kubectl delete...

But when we are talking about set replicas==0, I don’t think it will be the same, right?

What K8s do behind the wall? He just stop serving?

5

replicas == 0 scales it down to 0, deleting the running pods.

6

So, all the prehook stop and so on will be called!?

Hmm… So I’ll need to check why my application is not being gracefully shutdown…

Thanks!

7

Correct - there’s nothing hidden - “delete pod” happens in both paths.

8

When we do --replicas=0 that means it will repeat or treat same as delete process?

9

It will delete the pod. There’s only one delete.

10

DELETE request ----send----- API server
API server ------set deletionTimestamp ------ Pod
Kubelet ----------observe this settings on-------- Pod
Kubelet started to terminate Pod.
I draw a deletion process arch. can you correct me is it right or may be need more integration added?