Pod deletion timestamp is resetted after preStop hook is executed

this looks like a bug to me, having deletionTimestamps that fluctuate is confusing at minimum, let me try to understand the timeline:

t0 create pod with finalizer
t1 delete pod  -> deletionTimestamp is equal to t1+gracePeriod
at t1 + sleepPeriod -> deletionTimestamp is updated and is equal to t1??


yep, that is confirmed with 
```yaml
apiVersion: v1
kind: Pod
metadata:
  name: prestop-example
  finalizers:
    - test/finalizer
spec:
  terminationGracePeriodSeconds: 30
  containers:
  - name: app-container
    image: registry.k8s.io/e2e-test-images/agnhost:2.48
    lifecycle:
      preStop:
        sleep:
          seconds: 5

```

after preStop timestamp goes back in time

```
$ date && kubectl get pods  prestop-example -o yaml | grep -i timestam
Tue Jun 10 11:37:31 AM UTC 2025
  creationTimestamp: "2025-06-10T11:36:20Z"
  deletionTimestamp: "2025-06-10T11:37:56Z"
$ date && kubectl get pods  prestop-example -o yaml | grep -i timestam
Tue Jun 10 11:37:32 AM UTC 2025
  creationTimestamp: "2025-06-10T11:36:20Z"
  deletionTimestamp: "2025-06-10T11:37:32Z"

```

it seems the kubelet sends a new DELETE after the prestop hook is executed

```
I0610 11:55:01.644079       1 httplog.go:134] "HTTP" verb="DELETE" URI="/api/v1/namespaces/default/pods/prestop-example" latency="5.239176ms" userAgent="kubectl/v1.33.1 (linux/amd64) kubernetes/8adc0f0" audit-ID="1b48906d-bda7-4599-8cad-7ff983291186" srcIP="192.168.8.1:54692" apf_pl="global-default" apf_fs="global-default" apf_iseats=1 apf_fseats=0 apf_additionalLatency="0s" apf_execution_time="4.622037ms" resp=200




I0610 11:55:06.982183       1 handler.go:152] kube-aggregator: DELETE "/api/v1/namespaces/default/pods/prestop-example" satisfied by nonGoRestful
I0610 11:55:06.982209       1 handler.go:142] kube-apiserver: DELETE "/api/v1/namespaces/default/pods/prestop-example" satisfied by gorestful with webservice /api/v1


```

_Originally posted by @aojea in https://github.com/kubernetes/kubernetes/issues/128642#issuecomment-2958830949_
            

The logic that sets the deletion timestamp considering the grace period comes from

https://github.com/kubernetes/kubernetes/blob/54fd2aaefd11e12a3ecb6d1a1326f04cdc8ea1a3/pkg/registry/core/pod/strategy.go#L136-L138

and is added in

https://github.com/kubernetes/kubernetes/blob/da502386e53c1be6d476af8352337d82deb470aa/staging/src/k8s.io/apiserver/pkg/registry/rest/delete.go#L154-L169

I didn;t have time to trace the preStop hook logic, but it seems on the experiments that it sends a new Delete request , the logic seems to start here

https://github.com/kubernetes/kubernetes/blob/da502386e53c1be6d476af8352337d82deb470aa/pkg/kubelet/kuberuntime/kuberuntime_container.go#L731


	// `CheckGracefulDelete` will be implemented by specific strategy
	if !gracefulStrategy.CheckGracefulDelete(ctx, obj, options) {
	return false, false, nil
	}

	if options.GracePeriodSeconds == nil {
	return false, false, errors.NewInternalError(fmt.Errorf("options.GracePeriodSeconds should not be nil"))
	}

	requestedDeletionTimestamp := metav1.NewTime(metav1Now().Add(time.Second * time.Duration(*options.GracePeriodSeconds)))
	objectMeta.SetDeletionTimestamp(&requestedDeletionTimestamp)
	objectMeta.SetDeletionGracePeriodSeconds(options.GracePeriodSeconds)
	// If it's the first graceful deletion we are going to set the DeletionTimestamp to non-nil.
	// Controllers of the object that's being deleted shouldn't take any nontrivial actions, hence its behavior changes.
	// Thus we need to bump object's Generation (if set). This handles generation bump during graceful deletion.
	// The bump for objects that don't support graceful deletion is handled in pkg/registry/generic/registry/store.go.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Pod deletion timestamp is resetted after preStop hook is executed #132205

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	// CheckGracefulDelete allows a pod to be gracefully deleted. It updates the DeleteOptions to
	// reflect the desired grace value.
	func (podStrategy) CheckGracefulDelete(ctx context.Context, obj runtime.Object, options *metav1.DeleteOptions) bool {

Pod deletion timestamp is resetted after preStop hook is executed #132205

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions