github/kubernetes
1
0
Fork 0
mirror of https://github.com/k3s-io/kubernetes.git synced 2025-07-31 15:25:57 +00:00
Code Issues Projects Releases Wiki Activity

Merge pull request #31917 from derekwaynecarr/delete-pods-last

Browse Source 
Automatic merge from submit-queue

Namespace controller deletes pods last

I think this fixes https://github.com/kubernetes/kubernetes/issues/29308 or at least helps further reduce the incidence.

This PR changes the order in which namespace controller prioritizes resources for deletion.  It deletes all resources before deleting pods.  The rationale for this change is to broadcast deletion of controllers that spawn pods first rather than trip those controllers up into thinking they should spawn more pods which would increase the risk of causing races with the `NamespaceLifecycle` admission plug-in.  Many of those controllers also are not rate-limited in the face of rejection, so rather than promote a situation where they are rejected, we promote a situation that removes those things first.
This commit is contained in:
Kubernetes Submit Queue 2016-09-02 00:50:02 -07:00 committed by GitHub
commit 43093609c1
1 changed files with 20 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
pkg/controller/namespace/namespace_controller_utils.go
View File

@ -18,6 +18,7 @@ package namespace
import (
"fmt"
"sort"
"time"
"k8s.io/kubernetes/pkg/api"
@ -318,6 +319,8 @@ func deleteAllContent(
estimate := int64(0)
glog.V(4).Infof("namespace controller - deleteAllContent - namespace: %s, gvrs: %v", namespace, groupVersionResources)
// iterate over each group version, and attempt to delete all of its resources
// we sort resources to delete in a priority order that deletes pods LAST
sort.Sort(sortableGroupVersionResources(groupVersionResources))
for _, gvr := range groupVersionResources {
gvrEstimate, err := deleteAllContentForGroupVersionResource(kubeClient, clientPool, opCache, gvr, namespace, namespaceDeletedAt)
if err != nil {
@ -453,3 +456,20 @@ func estimateGracefulTerminationForPods(kubeClient clientset.Interface, ns strin
}
return estimate, nil
}
// sortableGroupVersionResources sorts the input set of resources for deletion, and orders pods to always be last.
// the idea is that the namespace controller will delete all things that spawn pods first in order to reduce the time
// those controllers spend creating pods only to be told NO in admission and potentially overwhelming cluster especially if they lack rate limiting.
type sortableGroupVersionResources []unversioned.GroupVersionResource
func (list sortableGroupVersionResources) Len() int {
return len(list)
}
func (list sortableGroupVersionResources) Swap(i, j int) {
list[i], list[j] = list[j], list[i]
}
func (list sortableGroupVersionResources) Less(i, j int) bool {
return list[j].Group == "" && list[j].Resource == "pods"
}