Doc Menu

Back up Data to S3-Compatible Storage Using Dumpling

This document describes how to back up the data of the TiDB cluster in Kubernetes to the S3-compatible storage. "Backup" in this document refers to full backup (ad-hoc full backup and scheduled full backup). For the underlying implementation, Dumpling is used to get the logic backup of the TiDB cluster, and then this backup data is sent to the S3-compatible storage.

The backup method described in this document is implemented based on CustomResourceDefinition (CRD) in TiDB Operator v1.1 or later versions. For the backup method implemented based on Helm Charts, refer to Back up and Restore TiDB Cluster Data Based on Helm Charts.

Ad-hoc full backup to S3-compatible storage

Ad-hoc full backup describes the backup by creating a Backup custom resource (CR) object. TiDB Operator performs the specific backup operation based on this Backup object. If an error occurs during the backup process, TiDB Operator does not retry and you need to handle this error manually.

For the current S3-compatible storage types, Ceph and Amazon S3 work normally as tested. Therefore, this document shows examples in which the data of the demo1 TiDB cluster in the test1 Kubernetes namespace is backed up to Ceph and Amazon S3 respectively.

Three methods to grant permissions of AWS account

  • If you use Amazon S3 to back up and restore the cluster, you have three methods to grant permissions. For details, refer to Back up TiDB Cluster Data to AWS Using BR.
  • If Ceph is used as backend storage in backup and restore test, the permission is granted by importing AccessKey and SecretKey.

Prerequisites for ad-hoc backup

Refer to Ad-hoc full backup prerequisites.

Required database account privileges

  • The SELECT and UPDATE privileges of the mysql.tidb table: Before and after the backup, the Backup CR needs a database account with these privileges to adjust the GC time.
  • SELECT
  • RELOAD
  • LOCK TABLES
  • REPLICATION CLIENT

Ad-hoc backup process

Note:

Because of the rclone issue, if the backup data is stored in Amazon S3 and the AWS-KMS encryption is enabled, you need to add the following spec.s3.options configuration to the YAML file in the examples of this section:

spec:
  ...
  s3:
    ...
    options:
    - --ignore-checksum

Examples:

  • Create the Backup CR, and back up cluster data to Amazon S3 by importing AccessKey and SecretKey to grant permissions:

    kubectl apply -f backup-s3.yaml

    The content of backup-s3.yaml is as follows:

    ---
    apiVersion: pingcap.com/v1alpha1
    kind: Backup
    metadata:
      name: demo1-backup-s3
      namespace: test1
    spec:
      from:
        host: ${tidb_host}
        port: ${tidb_port}
        user: ${tidb_user}
        secretName: backup-demo1-tidb-secret
      s3:
        provider: aws
        secretName: s3-secret
        region: ${region}
        bucket: ${bucket}
        # prefix: ${prefix}
        # storageClass: STANDARD_IA
        # acl: private
        # endpoint:
    # dumpling:
    #  options:
    #  - --threads=16
    #  - --rows=10000
    #  tableFilter:
    #  - "test.*"
      # storageClassName: local-storage
      storageSize: 10Gi
  • Create the Backup CR, and back up data to Ceph by importing AccessKey and SecretKey to grant permissions:

    kubectl apply -f backup-s3.yaml

    The content of backup-s3.yaml is as follows:

    ---
    apiVersion: pingcap.com/v1alpha1
    kind: Backup
    metadata:
      name: demo1-backup-s3
      namespace: test1
    spec:
      from:
        host: ${tidb_host}
        port: ${tidb_port}
        user: ${tidb_user}
        secretName: backup-demo1-tidb-secret
      s3:
        provider: ceph
        secretName: s3-secret
        endpoint: ${endpoint}
        # prefix: ${prefix}
        bucket: ${bucket}
    # dumpling:
    #  options:
    #  - --threads=16
    #  - --rows=10000
    #  tableFilter:
    #  - "test.*"
      # storageClassName: local-storage
      storageSize: 10Gi
  • Create the Backup CR, and back up data to Amazon S3 by binding IAM with Pod to grant permissions:

    kubectl apply -f backup-s3.yaml

    The content of backup-s3.yaml is as follows:

    ---
    apiVersion: pingcap.com/v1alpha1
    kind: Backup
    metadata:
    name: demo1-backup-s3
    namespace: test1
    annotations:
        iam.amazonaws.com/role: arn:aws:iam::123456789012:role/user
    spec:
    backupType: full
    from:
        host: ${tidb_host}
        port: ${tidb_port}
        user: ${tidb_user}
        secretName: backup-demo1-tidb-secret
    s3:
        provider: aws
        region: ${region}
        bucket: ${bucket}
        # prefix: ${prefix}
        # storageClass: STANDARD_IA
        # acl: private
        # endpoint:
    # dumpling:
    #  options:
    #  - --threads=16
    #  - --rows=10000
    #  tableFilter:
    #  - "test.*"
      # storageClassName: local-storage
      storageSize: 10Gi
  • Create the Backup CR, and back up data to Amazon S3 by binding IAM with ServiceAccount to grant permissions:

    kubectl apply -f backup-s3.yaml

    The content of backup-s3.yaml is as follows:

    ---
    apiVersion: pingcap.com/v1alpha1
    kind: Backup
    metadata:
    name: demo1-backup-s3
    namespace: test1
    spec:
    backupType: full
    serviceAccount: tidb-backup-manager
    from:
        host: ${tidb_host}
        port: ${tidb_port}
        user: ${tidb_user}
        secretName: backup-demo1-tidb-secret
    s3:
        provider: aws
        region: ${region}
        bucket: ${bucket}
        # prefix: ${prefix}
        # storageClass: STANDARD_IA
        # acl: private
        # endpoint:
    # dumpling:
    #  options:
    #  - --threads=16
    #  - --rows=10000
    #  tableFilter:
    #  - "test.*"
      # storageClassName: local-storage
      storageSize: 10Gi

In the examples above, all data of the TiDB cluster is exported and backed up to Amazon S3 and Ceph respectively. You can ignore the acl, endpoint, and storageClass configuration items in the Amazon S3 configuration. S3-compatible storage types other than Amazon S3 can also use a configuration similar to that of Amazon S3. You can also leave the configuration item fields empty if you do not need to configure these items as shown in the above Ceph configuration.

Amazon S3 supports the following ACL polices:

  • private
  • public-read
  • public-read-write
  • authenticated-read
  • bucket-owner-read
  • bucket-owner-full-control

If the ACL policy is not configured, the private policy is used by default. For the detailed description of these access control policies, refer to AWS documentation.

Amazon S3 supports the following storageClass types:

  • STANDARD
  • REDUCED_REDUNDANCY
  • STANDARD_IA
  • ONEZONE_IA
  • GLACIER
  • DEEP_ARCHIVE

If storageClass is not configured, STANDARD_IA is used by default. For the detailed description of these storage types, refer to AWS documentation.

After creating the Backup CR, you can use the following command to check the backup status:

kubectl get bk -n test1 -owide

More Backup CR parameter description:

  • .spec.metadata.namespace: the namespace where the Backup CR is located.

  • .spec.tikvGCLifeTime: the temporary tikv_gc_lifetime time setting during the backup. Defaults to 72h.

    Before the backup begins, if the tikv_gc_lifetime setting in the TiDB cluster is smaller than spec.tikvGCLifeTime set by the user, TiDB Operator adjusts the value of tikv_gc_lifetime to the value of spec.tikvGCLifeTime. This operation makes sure that the backup data is not garbage-collected by TiKV.

    After the backup, no matter whether the backup is successful or not, as long as the previous tikv_gc_lifetime is smaller than .spec.tikvGCLifeTime, TiDB Operator will try to set tikv_gc_lifetime to the previous value.

    In extreme cases, if TiDB Operator fails to access the database, TiDB Operator cannot automatically recover the value of tikv_gc_lifetime and treats the backup as failed. At this time, you can view tikv_gc_lifetime of the current TiDB cluster using the following statement:

    select VARIABLE_NAME, VARIABLE_VALUE from mysql.tidb where VARIABLE_NAME like "tikv_gc_life_time";

    In the output of the command above, if the value of tikv_gc_lifetime is still larger than expected (10m by default), it means TiDB Operator failed to automatically recover the value. Therefore, you need to set tikv_gc_lifetime back to the previous value manually:

    update mysql.tidb set VARIABLE_VALUE = '10m' where VARIABLE_NAME = 'tikv_gc_life_time';
    
  • .spec.cleanPolicy: The clean policy of the backup data when the backup CR is deleted.

    Three clean policies are supported:

    • Retain: On any circumstances, retain the backup data when deleting the backup CR.

    • Delete: On any circumstances, delete the backup data when deleting the backup CR.

    • OnFailure: If the backup fails, delete the backup data when deleting the backup CR.

      If this field is not configured, or if you configure a value other than the three policies above, the backup data is retained.

      Note that in v1.1.2 and earlier versions, this field does not exist. The backup data is deleted along with the CR by default. For v1.1.3 or later versions, if you want to keep this behavior, set this field to Delete.

  • .spec.from.host: the address of the TiDB cluster to be backed up, which is the service name of the TiDB cluster to be exported, such as basic-tidb.

  • .spec.from.port: the port of the TiDB cluster to be backed up.

  • .spec.from.user: the accessing user of the TiDB cluster to be backed up.

  • .spec.from.secretName:the secret contains the password of the .spec.from.user.

  • spec.s3.region: configures the Region where Amazon S3 is located if you want to use Amazon S3 for backup storage.

  • .spec.s3.bucket: the name of the bucket compatible with S3 storage.

  • .spec.s3.prefix: this field can be ignored. If you set this field, it will be used to make up the remote storage path s3://${.spec.s3.bucket}/${.spec.s3.prefix}/backupName.

  • .spec.dumpling: Dumpling-related configurations. You can specify Dumpling's operation parameters in the options field. See Dumpling Option list for more information. These configuration items of Dumpling can be ignored by default. When these items are not specified, the default values of options fields are as follows:

    options:
    - --threads=16
    - --rows=10000
  • .spec.storageClassName: the persistent volume (PV) type specified for the backup operation.

  • .spec.storageSize: the PV size specified for the backup operation (100 Gi by default). This value must be greater than the size of the TiDB cluster to be backed up.

    The PVC name corresponding to the Backup CR of a TiDB cluster is fixed. If the PVC already exists in the cluster namespace and the size is smaller than spec.storageSize, you need to delete this PVC and then run the Backup job.

  • .spec.tableFilter: Dumpling only backs up tables that match the table filter rules. This field can be ignored by default. If the field is not configured, the default value of tableFilter is as follows:

    tableFilter:
    - "*.*"
    - "!/^(mysql|test|INFORMATION_SCHEMA|PERFORMANCE_SCHEMA|METRICS_SCHEMA|INSPECTION_SCHEMA)$/.*"

    Note:

    To use the table filter to exclude db.table, you need to add the *.* rule to include all tables first. For example:

    tableFilter:
    - "*.*"
    - "!db.table"

Supported S3-compatible providers are as follows:

  • alibaba:Alibaba Cloud Object Storage System (OSS), formerly Aliyun
  • digitalocean:Digital Ocean Spaces
  • dreamhost:Dreamhost DreamObjects
  • ibmcos:IBM COS S3
  • minio:Minio Object Storage
  • netease:Netease Object Storage (NOS)
  • wasabi:Wasabi Object Storage
  • other:Any other S3 compatible provider

Scheduled full backup to S3-compatible storage

You can set a backup policy to perform scheduled backups of the TiDB cluster, and set a backup retention policy to avoid excessive backup items. A scheduled full backup is described by a custom BackupSchedule CR object. A full backup is triggered at each backup time point. Its underlying implementation is the ad-hoc full backup.

Prerequisites for scheduled backup

The prerequisites for the scheduled backup is the same as the prerequisites for ad-hoc backup.

Scheduled backup process

Note:

Because of the rclone issue, if the backup data is stored in Amazon S3 and the AWS-KMS encryption is enabled, you need to add the following spec.backupTemplate.s3.options configuration to the YAML file in the examples of this section:

spec:
  ...
  backupTemplate:
    ...
    s3:
      ...
      options:
      - --ignore-checksum

Examples:

  • Create the BackupSchedule CR to enable the scheduled full backup to Amazon S3 by importing AccessKey and SecretKey to grant permissions:

    kubectl apply -f backup-schedule-s3.yaml

    The content of backup-schedule-s3.yaml is as follows:

    ---
    apiVersion: pingcap.com/v1alpha1
    kind: BackupSchedule
    metadata:
      name: demo1-backup-schedule-s3
      namespace: test1
    spec:
      #maxBackups: 5
      #pause: true
      maxReservedTime: "3h"
      schedule: "*/2 * * * *"
      backupTemplate:
        from:
          host: ${tidb_host}
          port: ${tidb_port}
          user: ${tidb_user}
          secretName: backup-demo1-tidb-secret
        s3:
          provider: aws
          secretName: s3-secret
          region: ${region}
          bucket: ${bucket}
          # prefix: ${prefix}
          # storageClass: STANDARD_IA
          # acl: private
          # endpoint:
      # dumpling:
      #  options:
      #  - --threads=16
      #  - --rows=10000
      #  tableFilter:
      #  - "test.*"
        # storageClassName: local-storage
        storageSize: 10Gi
  • Create the BackupSchedule CR to enable the scheduled full backup to Ceph by importing AccessKey and SecretKey to grant permissions:

    kubectl apply -f backup-schedule-s3.yaml

    The content of backup-schedule-s3.yaml is as follows:

    ---
    apiVersion: pingcap.com/v1alpha1
    kind: BackupSchedule
    metadata:
      name: demo1-backup-schedule-ceph
      namespace: test1
    spec:
      #maxBackups: 5
      #pause: true
      maxReservedTime: "3h"
      schedule: "*/2 * * * *"
      backupTemplate:
        from:
          host: ${tidb_host}
          port: ${tidb_port}
          user: ${tidb_user}
          secretName: backup-demo1-tidb-secret
        s3:
          provider: ceph
          secretName: s3-secret
          endpoint: ${endpoint}
          bucket: ${bucket}
          # prefix: ${prefix}
      # dumpling:
      #  options:
      #  - --threads=16
      #  - --rows=10000
      #  tableFilter:
      #  - "test.*"
        # storageClassName: local-storage
        storageSize: 10Gi
  • Create the BackupSchedule CR to enable the scheduled full backup, and back up the cluster data to Amazon S3 by binding IAM with Pod to grant permissions:

    kubectl apply -f backup-schedule-s3.yaml

    The content of backup-schedule-s3.yaml is as follows:

    ---
    apiVersion: pingcap.com/v1alpha1
    kind: BackupSchedule
    metadata:
      name: demo1-backup-schedule-s3
      namespace: test1
      annotations:
        iam.amazonaws.com/role: arn:aws:iam::123456789012:role/user
    spec:
      #maxBackups: 5
      #pause: true
      maxReservedTime: "3h"
      schedule: "*/2 * * * *"
      backupTemplate:
        from:
          host: ${tidb_host}
          port: ${tidb_port}
          user: ${tidb_user}
          secretName: backup-demo1-tidb-secret
        s3:
          provider: aws
          region: ${region}
          bucket: ${bucket}
          # prefix: ${prefix}
          # storageClass: STANDARD_IA
          # acl: private
          # endpoint:
      # dumpling:
      #  options:
      #  - --threads=16
      #  - --rows=10000
      #  tableFilter:
      #  - "test.*"
        # storageClassName: local-storage
        storageSize: 10Gi
  • Create the BackupSchedule CR to enable the scheduled full backup, and back up the cluster data to Amazon S3 by binding IAM with ServiceAccount to grant permissions:

    kubectl apply -f backup-schedule-s3.yaml

    The content of backup-schedule-s3.yaml is as follows:

    ---
    apiVersion: pingcap.com/v1alpha1
    kind: BackupSchedule
    metadata:
      name: demo1-backup-schedule-s3
      namespace: test1
    spec:
      #maxBackups: 5
      #pause: true
      maxReservedTime: "3h"
      schedule: "*/2 * * * *"
      serviceAccount: tidb-backup-manager
      backupTemplate:
        from:
          host: ${tidb_host}
          port: ${tidb_port}
          user: ${tidb_user}
          secretName: backup-demo1-tidb-secret
        s3:
          provider: aws
          region: ${region}
          bucket: ${bucket}
          # prefix: ${prefix}
          # storageClass: STANDARD_IA
          # acl: private
          # endpoint:
      # dumpling:
      #  options:
      #  - --threads=16
      #  - --rows=10000
      #  tableFilter:
      #  - "test.*"
        # storageClassName: local-storage
        storageSize: 10Gi

After creating the scheduled full backup, you can use the following command to check the backup status:

kubectl get bks -n test1 -owide

You can use the following command to check all the backup items:

kubectl get bk -l tidb.pingcap.com/backup-schedule=demo1-backup-schedule-s3 -n test1

From the examples above, you can see that the backupSchedule configuration consists of two parts. One is the unique configuration of backupSchedule, and the other is backupTemplate. backupTemplate specifies the configuration related to the S3-compatible storage, which is the same as the configuration of the ad-hoc full backup to the S3-compatible storage (refer to Ad-hoc backup process for details). The following are the unique configuration items of backupSchedule:

  • .spec.maxBackups: A backup retention policy, which determines the maximum number of backup items to be retained. When this value is exceeded, the outdated backup items will be deleted. If you set this configuration item to 0, all backup items are retained.
  • .spec.maxReservedTime: A backup retention policy based on time. For example, if you set the value of this configuration to 24h, only backup items within the recent 24 hours are retained. All backup items out of this time are deleted. For the time format, refer to func ParseDuration. If you have set the maximum number of backup items and the longest retention time of backup items at the same time, the latter setting takes effect.
  • .spec.schedule: The time scheduling format of Cron. Refer to Cron for details.
  • .spec.pause: false by default. If this parameter is set to true, the scheduled scheduling is paused. In this situation, the backup operation will not be performed even if the scheduling time is reached. During this pause, the backup Garbage Collection (GC) runs normally. If you change true to false, the full backup process is restarted.

Note:

TiDB Operator creates a PVC. This PVC is used for both ad-hoc full backup and scheduled full backup. The backup data is stored in PV first and then uploaded to remote storage. If you want to delete this PVC after the backup is completed, you can refer to Delete Resource to delete the backup Pod first, and then delete the PVC.

If the backup data is successfully uploaded to remote storage, TiDB Operator automatically deletes the local data. If the upload fails, the local data is retained.

Delete the backup CR

You can delete the full backup CR (Backup) and the scheduled backup CR (BackupSchedule) by the following commands:

kubectl delete backup ${name} -n ${namespace}
kubectl delete backupschedule ${name} -n ${namespace}

If you use TiDB Operator v1.1.2 or earlier versions, or if you use TiDB Operator v1.1.3 or later versions and set the value of spec.cleanPolicy to Delete, TiDB Operator deletes the backup data when it deletes the CR. In such cases, if you need to delete the namespace as well, it is recommended that you first delete all the Backup/BackupSchedule CR and then delete the namespace.

If you delete the namespace before you delete the Backup/BackupSchedule CR, TiDB Operator continues to create jobs to clean the backup data. However, since the namespace is in Terminating state, TiDB Operator fails to create a job, which causes the namespace to be stuck in the state.

For v1.1.2 and earlier versions, if the backup data is manually deleted before you delete the Backup/BackupSchedule CR, the namespace might also be stuck in the Terminating state.

To address this issue, delete finalizers using the following command:

kubectl edit backup ${name} -n ${namespace}

After deleting the metadata.finalizers configuration, you can delete CR normally.

Troubleshooting

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