tidb-google-maintenance

We use similar approach as aerospike: watch GCP maintenance events on TiDB/TiKV/PD nodes and take proper actions:

• TiDB: Put the TiDB offline by cordon the TiDB node and delete the TiDB pod (the node pool of TiDB instance MUST be set to auto-scale, the cordon node is expected to be reclaimed by auto-scaler)
• TiKV: Ecivt leaders on TiKV store during maintenance.
• PD: Resign leader if the current PD instance is the PD leader

An additional container is added to run the maintenance watching script.

Deploy

Sidecar Image

Used Sidecar Public image: pingcap/tidb-gcp-live-migration:${TIDB_VERSION} (e.g. pingcap/tidb-gcp-live-migration:v7.1.0)

or build the image on your own: TIDB_VERSION=v7.1.0 IMAGE=${YOUR_IMAGE}/tidb-gcp-live-migration make image-release

Add the Sidecar Image into manifest

For Cluster with TLS Enabled

For TiDB, add content below to spec.tidb (replace ${CLUSTR_NAME})

run

# replace ${SERVICEACCOUNT}, ${NAMESPACE} and ${CLUSTR_NAME}
kubectl apply -f rbac.yaml

        additionalContainers:
          - command:
              - python3
              - /main.py
            env:
              - name: TLS
                value: true
              - name: CLUSTER_NAME
                value: ${CLUSTR_NAME}
              - name: ROLE
                value: tidb
              - name: NODENAME
                valueFrom:
                  fieldRef:
                    fieldPath: spec.nodeName
            image: pingcap/tidb-gcp-live-migration:v7.1.0
            name: gcp-maintenance-script

For TiKV, add content below to spec.tikv (replace ${CLUSTR_NAME})

        additionalVolumes:
          - name: pd-tls
            secret:
              secretName: ${CLUSTR_NAME}-pd-cluster-secret
        additionalContainers:
          - command:
              - python3
              - /main.py
            env:
              - name: TLS
                value: true
              - name: CLUSTER_NAME
                value: ${CLUSTR_NAME}
              - name: ROLE
                value: tikv
            image: pingcap/tidb-gcp-live-migration:v7.1.0
            name: gcp-maintenance-script
            volumeMounts:
              - name: pd-tls
                mountPath: /var/lib/pd-tls
              - name: tikv-tls
                mountPath: /var/lib/tikv-tls

For PD, add content below to spec.pd (replace ${CLUSTR_NAME}),

        additionalContainers:
          - command:
              - python3
              - /main.py
            env:
              - name: TLS
                value: true
              - name: CLUSTER_NAME
                value: ${CLUSTR_NAME}
              - name: ROLE
                value: PD
            image: pingcap/tidb-gcp-live-migration:v7.1.0
            name: gcp-maintenance-script
            volumeMounts:
              - name: pd-tls
                mountPath: /var/lib/pd-tls

For Cluster with TLS Disabled

For TiDB, add content below to spec.tidb (replace ${CLUSTR_NAME})

run

# replace ${SERVICEACCOUNT}, ${NAMESPACE} and ${CLUSTR_NAME}
kubectl apply -f rbac.yaml

        additionalContainers:
          - command:
              - python3
              - /main.py
            env:
              - name: TLS
                value: false
              - name: CLUSTER_NAME
                value: ${CLUSTR_NAME}
              - name: ROLE
                value: tidb
              - name: NODENAME
                valueFrom:
                  fieldRef:
                    fieldPath: spec.nodeName
            image: pingcap/tidb-gcp-live-migration:v7.1.0
            name: gcp-maintenance-script

For TiKV, add content below to spec.tikv (replace ${CLUSTR_NAME})

        additionalContainers:
          - command:
              - python3
              - /main.py
            env:
              - name: TLS
                value: false
              - name: CLUSTER_NAME
                value: ${CLUSTR_NAME}
              - name: ROLE
                value: tikv
            image: pingcap/tidb-gcp-live-migration:v7.1.0
            name: gcp-maintenance-script

For PD, add content below to spec.pd (replace ${CLUSTR_NAME}),

        additionalContainers:
          - command:
              - python3
              - /main.py
            env:
              - name: TLS
                value: false
              - name: CLUSTER_NAME
                value: ${CLUSTR_NAME}
              - name: ROLE
                value: PD
            image: pingcap/tidb-gcp-live-migration:v7.1.0
            name: gcp-maintenance-script

PD scheduler configuration

Increase the PD leader-schedule limit after the cluster is deployed, through sql:

set config pd `leader-schedule-limit`=100;

To Simulate a GCP Maintenance Event

see: https://cloud.google.com/compute/docs/instances/simulating-host-maintenance

And a script to simulate the effect when live migration happened.