How to remove Kubernetes 'shutdown' pods

Question

I recently noticed a big accumulation of pods with status 'Shutdown'. We have been using Kubernetes since October, 2020.

Production and staging is running on the same nodes except that staging uses preemtible nodes to cut the cost. The containers are also stable in staging. (Failures occur rarely as they are caught in testing before).

Service provider Google Cloud Kubernetes.

I familiarized myself with the docs and tried searching however neither I recognize neither google helps with this particular status. There are no errors in the logs.

I have no problem pods being stopped. Ideally I'd like K8s to automatically delete these shutdown pods. If I run kubectl delete po redis-7b86cdccf9-zl6k9 , it goes away in a blink.

kubectl get pods | grep Shutdown | awk '{print $1}' | xargs kubectl delete pod kubectl get pods | grep Shutdown | awk '{print $1}' | xargs kubectl delete pod is manual temporary workaround.

PS. k is an alias to kubectl in my environment.

Final example: it happens across all namespaces // different containers.

I stumbled upon few related issues explaining the status https://github.com/kubernetes/website/pull/28235 https://github.com/kubernetes/kubernetes/issues/102820

"When pods were evicted during the graceful node shutdown, they are marked as failed. Running kubectl get pods shows the status of the the evicted pods as Shutdown ."

Answer 1

The evicted pods are not removed on purpose, as k8s team says here 1 , the evicted pods are nor removed in order to be inspected after eviction.

I believe here the best approach would be to create a cronjob 2 as already mentioned.

apiVersion: batch/v1
kind: CronJob
metadata:
  name: del-shutdown-pods
spec:
  schedule: "* 12 * * *"
  jobTemplate:
    spec:
      template:
        spec:
          containers:
          - name: hello
            image: busybox
            imagePullPolicy: IfNotPresent
            command:
            - /bin/sh
            - -c
            - kubectl get pods | grep Shutdown | awk '{print $1}' | xargs kubectl delete pod
          restartPolicy: OnFailure

Answer 2

You don't need any grep - just use selectors that kubectl provides. And, btw, you cannot call kubectl from the busybox image, because it doesn't have kubectl at all. I also created a service account with the right of pods deletion.

apiVersion: batch/v1
kind: CronJob
metadata:
  name: del-shutdown-pods
spec:
  schedule: "0 */2 * * *"  
  concurrencyPolicy: Replace
  jobTemplate:
    metadata:
      name: shutdown-deleter
    spec:
      template:
        spec:
          serviceAccountName: deleter
          containers:
          - name: shutdown-deleter
            image: bitnami/kubectl
            imagePullPolicy: IfNotPresent
            command:
              - "/bin/sh"
            args:
              - "-c"
              - "kubectl delete pods --field-selector status.phase=Failed -A --ignore-not-found=true"
          restartPolicy: Never

Answer 3

First, try to forcefully delete the kubernetes pod using the below command:

$ kubectl delete pod <pod_name> -n --grace-period 0 --force

You can directly delete the pod using the below command:

$ kubectl delete pod

Then, check the status of the pod using the below command:

$ kubectl get pods

Here, you will see that pods have been deleted.

You can also verify using the documentation in the yaml file also.

Most programs gracefully shut down when receiving a SIGTERM, but if you are using third-party code or are managing a system you don't have control over, the preStop hook is a great way to trigger a graceful shutdown without modifying the application. Kubernetes will send a SIGTERM signal to the containers in the pod. At this point, Kubernetes waits for a specified time called the termination grace period.

For more information refer .

Answer 4

Right now Kubernetes doesn't remove evicted and shutdown status pods by default. We also faced a similar kind of issue in our environment.

As an automatic fix, you can create a Kubernetes cronjob which can remove the pod with evicted and shutdown status. The Kubernetes cronjob pod can authenticate using the serviceaccount and RBAC where you can restrict the verbs and namespaces for your utility.

Answer 5

您可以使用https://github.com/hjacobs/kube-janitor 。这提供了各种可配置的选项来清理

Answer 6

My take on this problem looks something like this (inspiration from other solutions here):

# Delete all shutdown pods. This is common problem on kubernetes using preemptible nodes on gke
# why awk, not failed pods: https://github.com/kubernetes/kubernetes/issues/54525#issuecomment-340035375
# due fact failed will delete evicted pods, that will complicate pod troubleshooting

---
apiVersion: batch/v1beta1
kind: CronJob
metadata:
  name: del-shutdown-pods
  namespace: kube-system
  labels:
    app: shutdown-pod-cleaner
spec:
  schedule: "*/1 * * * *"
  successfulJobsHistoryLimit: 1
  jobTemplate:
    spec:
      template:
        metadata:
          labels:
            app: shutdown-pod-cleaner
        spec:
          volumes:
          - name: scripts
            configMap:
              name: shutdown-pods-scripts
              defaultMode: 0777
          serviceAccountName: shutdown-pod-sa
          containers:
          - name: zombie-killer
            image: bitnami/kubectl
            imagePullPolicy: IfNotPresent
            command:
              - "/bin/sh"
            args:
              - "-c"
              - "/scripts/podCleaner.sh"
            volumeMounts:
              - name: scripts
                mountPath: "/scripts"
                readOnly: true
          restartPolicy: OnFailure
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: shutdown-pod-cleaner
  namespace: kube-system
  labels:
    app: shutdown-pod-cleaner
rules:
- apiGroups: [""]
  resources: ["pods"]
  verbs: ["delete", "list"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: shutdown-pod-cleaner-cluster
  namespace: kube-system
subjects:
- kind: ServiceAccount
  name: shutdown-pod-sa
  namespace: kube-system
roleRef:
  kind: ClusterRole
  name: shutdown-pod-cleaner
  apiGroup: ""
---
apiVersion: v1
kind: ServiceAccount
metadata:
  name: shutdown-pod-sa
  namespace: kube-system
  labels:
    app: shutdown-pod-cleaner
---
apiVersion: v1
kind: ConfigMap
metadata:
  name: shutdown-pods-scripts
  namespace: kube-system
  labels:
    app: shutdown-pod-cleaner
data:
  podCleaner.sh: |
    #!/bin/sh
    if [ $(kubectl get pods --all-namespaces --ignore-not-found=true | grep Shutdown | wc -l) -ge 1 ]
    then
    kubectl get pods -A | grep Shutdown | awk '{print $1,$2}' | xargs -n2 sh -c 'kubectl delete pod -n $0 $1 --ignore-not-found=true'
    else
    echo "no shutdown pods to clean"
    fi

Answer 7

I just set up a cronjob to clean the dead GKE pods. Complete setup includes RBAC role, role binding, and a service account.

Service account and cluster role setup.

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: pod-accessor-role
rules:
- apiGroups: [""]
  resources: ["pods"]
  verbs: ["get", "delete", "watch", "list"]
---

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: pod-access
subjects:
- kind: ServiceAccount
  name: cronjob-svc
  namespace: kube-system
roleRef:
  kind: ClusterRole
  name: pod-accessor-role
  apiGroup: rbac.authorization.k8s.io

---
apiVersion: v1
kind: ServiceAccount
metadata:
  name: cronjob-svc
  namespace: kube-system

Cronjob to clean up dead pods.

apiVersion: batch/v1
kind: CronJob
metadata:
  name: pod-cleaner-cron
  namespace: kube-system
spec:
  schedule: "0 */12 * * *"
  successfulJobsHistoryLimit: 1
  jobTemplate:
    spec:
      template:
        metadata:
          name: pod-cleaner-cron
          namespace: kube-system
        spec:
          serviceAccountName: cronjob-svc
          restartPolicy: Never
          containers:
          - name: pod-cleaner-cron
            imagePullPolicy: IfNotPresent
            image: bitnami/kubectl
            command:
              - "/bin/sh"
            args:
              - "-c"
              - "kubectl delete pods --field-selector status.phase=Failed -A --ignore-not-found=true"
status: {}