Force Delete Namespace in Kubernetes

Do you have a namespace in your Kubernetes cluster stuck in “Terminating” state? When a namespace is stuck in this state it usually means Kubernetes is not successful in cleaning up resources in that namespace. The main reason is finalizers or resources are not properly cleaned up or stuck. See example below:

$  kubectl get ns
NAME               STATUS        AGE
calico-apiserver   Active        4d13h
calico-system      Active        4d13h
default            Active        4d13h
kube-node-lease    Active        4d13h
kube-public        Active        4d13h
kube-system        Active        4d13h
longhorn-system    Terminating   18m
tigera-operator    Active        4d13h

I can confirm the pods and other resources are deleted but the state remains “Terminating“. Let’s consider few ways we can force delete his namespace in Kubernetes.

Set the stuck namespace name in NS variable.

NS=longhorn-system

Before forcing the deletion of the namespace, check for any pending resources that may be causing issues:

kubectl get all --namespace=$NS

1. Unset finalizers manually

If a namespace have finalizers not being properly removed, it can be stuck in “terminating” state. You can edit the namespace object yaml file:

kubectl edit namespace $NS

In the editor, locate finalizers field and delete by setting as below:

finalizers: []

2. Force Delete Using kubectl

If the first option is not successful, you can force namespace deletion by patching it.

Force deletion of the stuck namespace in kubernetes.

kubectl get $NS  -o json | jq '.spec.finalizers=[]' | kubectl replace --raw "/api/v1/ $NS/longhorn-system/finalize" -f -

Note: jq should be installed to enable you to modify JSON output. Be informed that this method bypasses the finalizers.

3. Forcefully delete the namespace

The--grace-period=0 and --force flags can be used to force namespace delete in Kubernetes. However, this method might leave orphaned resources and it should be used with caution:

kubectl delete namespace $NS --grace-period=0 --force

You can confirm if the namespace has been deleted by running the command:

kubectl get $NS

The output below shows it doesn’t exist:

error: the server doesn't have a resource type "longhorn-system"