Restart a TiDB Cluster in Kubernetes

This document describes how to forcibly restart a TiDB cluster in the Kubernetes cluster, including restarting a Pod, restarting all Pods of a component, and restarting all Pods of the TiDB cluster.

Note

TiDB Operator v1.0.x only supports restarting the Pod in a forcible way.

If the PD Pod being restarted is the PD leader, its leadership does not automatically transfer, which can cause the PD service to be temporarily unavailable.
If the TiKV Pod being restarted contains a Region leader of the TiKV cluster, its leadership does not automatically transfer, which can cause the requests of accessing the corresponding data to fail.
Restarting the TiDB Pod can cause the requests of accessing this Pod to fail.

Forcibly restart a Pod

Execute the following command to forcibly restart a Pod:

kubectl delete pod -n <namespace> <pod-name>

Forcibly restart all Pods of a component

Execute the following command to check the Pod list of a component:

kubectl get pod -n <namespace> -l app.kubernetes.io/component=<component-name>

Execute the following command to forcibly restart all Pods of a component:

kubectl delete pod -n <namespace> -l app.kubernetes.io/component=<component-name>

Note

To forcibly restart all Pods of the PD, TiDB, or TiKV component, replace <component-name> in the above commands with pd, tidb, or tikv respectively.
Replace <namespace> in the above commands with your own namespace.

Forcibly restart all Pods of the TiDB cluster

Execute the following command to check the Pod list of the TiDB cluster (including monitor and discovery):

kubectl get pod -n <namespace> -l  app.kubernetes.io/instance=<tidb-cluster-name>

Execute the following command to forcibly restart all Pods of the TiDB cluster (including monitor and discovery):

kubectl delete pod -n <namespace> -l  app.kubernetes.io/instance=<tidb-cluster-name>

Note

Replace <namespace> and <tidb-cluster-name> in the above commands with your own namespace and TiDB cluster name.