Suspend and Resume a TiDB Cluster on Kubernetes

This document describes how to suspend and resume a TiDB cluster on Kubernetes by configuring the Cluster object. When you suspend a cluster, all component Pods are stopped, but the Cluster object and associated resources (such as Services and PVCs) are retained. This preserves the cluster's data and configuration for later recovery.

Usage scenarios

You can suspend a TiDB cluster in the following scenarios:

• Temporarily release compute resources in a test environment.
• Stop a development cluster that is unused for an extended period.
• Pause a cluster temporarily while retaining its data and configuration.

Before you begin

Before you suspend a TiDB cluster, consider the following:

• Suspending the cluster interrupts cluster services.
• Existing client connections are terminated forcefully.
• PVCs and data are retained and continue to occupy storage space.
• Services and configurations associated with the cluster remain unchanged.

Suspend a TiDB cluster

To suspend a TiDB cluster, perform the following steps:

1. In the Cluster object, set the spec.suspendAction.suspendCompute field to true to suspend the entire TiDB cluster:

   apiVersion: core.pingcap.com/v1alpha1
   kind: Cluster
   metadata:
     name: ${cluster_name}
     namespace: ${namespace}
   spec:
     suspendAction:
       suspendCompute: true
     # ...
   

2. After the cluster is suspended, run the following command to observe the Pods being gradually deleted:

   kubectl -n ${namespace} get pods -w
   

Resume a TiDB cluster

To resume a suspended TiDB cluster, perform the following steps:

1. In the Cluster object, set the spec.suspendAction.suspendCompute field to false to resume the suspended TiDB cluster:

   apiVersion: core.pingcap.com/v1alpha1
   kind: Cluster
   metadata:
     name: ${cluster_name}
     namespace: ${namespace}
   spec:
     suspendAction:
       suspendCompute: false
     # ...
   

2. After the cluster is resumed, run the following command to observe the Pods being gradually created:

   kubectl -n ${namespace} get pods -w