Dong, Qiu Jian
f7b9a88535
Reviewed-by: Kovács, Zoltán <zkovacs@t-systems.com> Co-authored-by: Dong, Qiu Jian <qiujiandong1@huawei.com> Co-committed-by: Dong, Qiu Jian <qiujiandong1@huawei.com>
5.8 KiB
How Do I Evict All Pods on a Node?
You can run the kubectl drain command to safely evict all pods from a node.
By default, the kubectl drain command retains some system pods, for example, everest-csi-driver.
- Use kubectl to connect to the cluster.
- Check the nodes in the cluster.
kubectl get node
- Select a node and view all pods on the node.
kubectl get pod --all-namespaces -owide --field-selector spec.nodeName=192.168.0.160
The pods on the node before eviction are as follows:
NAMESPACE NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES default nginx-5bcc57c74b-lgcvh 1/1 Running 0 7m25s 10.0.0.140 192.168.0.160 <none> <none> kube-system coredns-6fcd88c4c-97p6s 1/1 Running 0 3h16m 10.0.0.138 192.168.0.160 <none> <none> kube-system everest-csi-controller-56796f47cc-99dtm 1/1 Running 0 3h16m 10.0.0.139 192.168.0.160 <none> <none> kube-system everest-csi-driver-dpfzl 2/2 Running 2 12d 192.168.0.160 192.168.0.160 <none> <none> kube-system icagent-tpfpv 1/1 Running 1 12d 192.168.0.160 192.168.0.160 <none> <none>
- Evict all pods on the node.
kubectl drain 192.168.0.160
If a pod mounted with local storage or controlled by a DaemonSet exists on the node, the message "eerror: unable to drain node "192.168.0.160" due to error: cannot delete DaemonSet-managed Pods..." will be displayed. The eviction command does not take effect. You can add the following parameters to the end of the preceding command to forcibly evict the pod:
- --delete-emptydir-data: forcibly evicts pods mounted with local storage, for example, coredns.
- --ignore-daemonsets: forcibly evicts the DaemonSet pods, for example, everest-csi-driver.
In the example, both types of pods exist on the node. Therefore, the eviction command is as follows:
kubectl drain 192.168.0.160 --delete-emptydir-data --ignore-daemonsets
- After the eviction, the node is automatically marked as unschedulable. That is, the node is tainted node.kubernetes.io/unschedulable = : NoSchedule.
After the eviction, only system pods are retained on the node.
NAMESPACE NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES kube-system everest-csi-driver-dpfzl 2/2 Running 2 12d 192.168.0.160 192.168.0.160 <none> <none> kube-system icagent-tpfpv 1/1 Running 1 12d 192.168.0.160 192.168.0.160 <none> <none>
Related Operations
Drain, cordon, and uncordon operations of kubectl:
- drain: Safely evicts all pods from a node and marks the node as unschedulable.
- cordon: Marks the node as unschedulable. That is, the node is tainted node.kubernetes.io/unschedulable = : NoSchedule.
- uncordon: Marks the node as schedulable.
For more information, see the kubectl documentation.