Restore Data from GCS Using BR

This document describes how to restore the TiDB cluster data backed up using TiDB Operator in Kubernetes.

The restore method described in this document is implemented based on CustomResourceDefinition (CRD) in TiDB Operator. For the underlying implementation, BR is used to restore the data. BR stands for Backup & Restore, which is a command-line tool for distributed backup and recovery of the TiDB cluster data.

Usage Scenarios

After backing up TiDB cluster data to GCS using BR, if you need to recover the backup SST (key-value pairs) files from GCS to TiDB cluster, you can follow steps in this document to restore the data using BR.

Restore Data Using BR - 图1Note

  • BR is only applicable to TiDB v3.1 or later releases.
  • Data restored by BR cannot be replicated to a downstream cluster, because BR directly imports SST files to TiDB and the downstream cluster currently cannot access the upstream SST files.

This document provides an example about how to restore the backup data from the spec.gcs.prefix folder of the spec.gcs.bucket bucket on GCS to the demo2 TiDB cluster in the test2 namespace. The following are the detailed steps.

Step 1: Prepare the restore environment

Before restoring backup data on GCS to TiDB using BR, take the following steps to prepare the restore environment:

  1. Download backup-rbac.yaml, and execute the following command to create the role-based access control (RBAC) resources in the test2 namespace:

    1. kubectl apply -f backup-rbac.yaml -n test2
  2. Grant permissions to the remote storage.

    Refer to GCS account permissions.

  3. For a TiDB version earlier than v4.0.8, you also need to complete the following preparation steps. For TiDB v4.0.8 or a later version, skip these preparation steps.

    1. Make sure that you have the SELECT and UPDATE privileges on the mysql.tidb table of the target database so that the Restore CR can adjust the GC time before and after the restore.

    2. Create the restore-demo2-tidb-secret secret to store the root account and password to access the TiDB cluster:

      1. kubectl create secret generic restore-demo2-tidb-secret --from-literal=user=root --from-literal=password=<password> --namespace=test2

Step 2: Restore the backup data to a TiDB cluster

  1. Create the Restore custom resource (CR) to restore the specified data to your cluster:

    1. kubectl apply -f restore.yaml

    The content of restore.yaml file is as follows:

    1. ---
    2. apiVersion: pingcap.com/v1alpha1
    3. kind: Restore
    4. metadata:
    5. name: demo2-restore-gcs
    6. namespace: test2
    7. spec:
    8. # backupType: full
    9. br:
    10. cluster: demo2
    11. clusterNamespace: test2
    12. # logLevel: info
    13. # statusAddr: ${status-addr}
    14. # concurrency: 4
    15. # rateLimit: 0
    16. # checksum: true
    17. # sendCredToTikv: true
    18. # Only needed for TiDB Operator < v1.1.10 or TiDB < v4.0.8
    19. # to:
    20. # host: ${tidb_host}
    21. # port: ${tidb_port}
    22. # user: ${tidb_user}
    23. # secretName: restore-demo2-tidb-secret
    24. gcs:
    25. projectId: ${project_id}
    26. secretName: gcs-secret
    27. bucket: ${bucket}
    28. prefix: ${prefix}
    29. # location: us-east1
    30. # storageClass: STANDARD_IA
    31. # objectAcl: private

    When configuring restore.yaml, note the following:

    • For more information about GCS configuration, refer to GCS fields.
    • Some parameters in .spec.br are optional, such as logLevel, statusAddr, concurrency, rateLimit, checksum, timeAgo, and sendCredToTikv. For more information about BR configuration, refer to BR fields.
    • For v4.0.8 or a later version, BR can automatically adjust tikv_gc_life_time. You do not need to configure spec.to fields in the Restore CR.
    • For more information about the Restore CR fields, refer to Restore CR fields.
  2. After creating the Restore CR, execute the following command to check the restore status:

    1. kubectl get rt -n test2 -owide

Troubleshooting

If you encounter any problem during the restore process, refer to Common Deployment Failures.