Back up and restore in Kubernetes

This page shows you how to back up and restore your AlloyDB Omni data using the AlloyDB Omni Kubernetes operator. This requires basic knowledge about updating a Kubernetes cluster using manifest files and the kubectl command-line tool. For more information about installing and running AlloyDB Omni on a Kubernetes cluster, see Install AlloyDB on Kubernetes.

To enable AlloyDB Omni continuous backup and recovery, you must create a backup plan for each database cluster. Backups are taken based on the backup schedules defined in the backupPlan resource. If no backup schedule is defined in the backup plan, continuous backups are taken daily by default. You can restore or clone backups from any timestamp in the recovery window at seconds granularity.

For information about backing up and restoring your AlloyDB Omni data on non-Kubernetes deployments, see Install and configure Barman.

Enable and schedule backups

Continuous backups are enabled when you create a backup plan resource for your database cluster. You must create a backupPlan resource for each database cluster to enable continuous backup for that cluster. This backup plan resource defines the following parameters:

  • The location that the AlloyDB Omni operator stores backups to. This can be local to your Kubernetes cluster, or to a Cloud Storage bucket.

  • An option to set multiple backup schedules that automatically create full, incremental, and differential backups. You can pause this schedule at any time, including when initially defining the backup plan. If a backup plan is paused, scheduled backups aren't created, but you can still use it to create backups manually.

    If no backup schedules are specified, then the default is "0 0 * * *", which takes one daily full backup at midnight, local time.

  • A retention period for stored backups. This can be as short as one day, or as long as 90 days. The default value is 14.

Your database cluster can have multiple backup plans, each with their own name and configuration. If you create multiple backupPlan resources with different backup schedules for a database cluster, then you must define a unique backup location for each backup resource.

Create a plan to store backups locally

To enable backups that get locally stored, apply the following manifest:

    apiVersion: alloydbomni.dbadmin.goog/v1
    kind: BackupPlan
    metadata:
      name: BACKUP_PLAN_NAME
      namespace: NAMESPACE
    spec:
      dbclusterRef: DB_CLUSTER_NAME
      backupSchedules:
        full: "FULL_CRON_SCHEDULE"
        differential: "DIFF_CRON_SCHEDULE"
        incremental: "INCR_CRON_SCHEDULE"
      backupRetainDays: RETENTION_DAYS
      paused: PAUSED_BOOLEAN

Replace the following:

  • BACKUP_PLAN_NAME: a name for this backup plan resource—for example, backup-plan-1.

  • NAMESPACE: the Kubernetes namespace for this backup plan. It must match the namespace of the database cluster.

  • DB_CLUSTER_NAME: the name of your database cluster, which you assigned when you created it.

  • FULL_CRON_SCHEDULE: a backup schedule to create a full backup, containing all data, expressed in cron format. For example, set to "0 0 * * 0" to take a full backup at 00:00 on every Sunday.

  • DIFF_CRON_SCHEDULE: a backup schedule to create backups that are initially full backups. Subsequent backups are differential, based on intervening changes to the data, expressed in cron format. For example, set to "0 22 * * 3"to take a differential backup at 22:00 on every Wednesday.

  • INCR_CRON_SCHEDULE: a backup schedule to create backups that include data that has changed from the last full, differential, or incremental backup. It is expressed in cron format. For example, set to "0 21 * * *" to take an incremental backup at 21:00 every day.

  • RETENTION_DAYS: the number of days for which the AlloyDB Omni Operator retains this backup. It must be an integer between 1 and 90. The default value is 14.

  • PAUSED_BOOLEAN: specifies whether or not the backup plan is paused. Provide one of the following values:

    • true: backups are paused and no scheduled backups are created.

    • false: the AlloyDB Omni Operator creates backups according to the schedule specified by cronSchedule. This is the default value, if not explicitly set to true.

    The default value is false.

Create a plan that stores backups in the Cloud Storage

To enable backups that get stored on Cloud Storage, follow these steps:

  1. Create a Cloud Storage bucket. Note the name that you assign to this bucket; you need it in a later step.

  2. Create a service account for adding backups to the bucket.

  3. Grant the storage.objectAdmin Identity and Access Management role to the service account.

  4. Create a key for the service account. This downloads the private key to your local environment.

  5. Rename the downloaded key file to key.json.

  6. Create a Kubernetes secret containing the private key:

    kubectl create secret generic SECRET_NAME --from-file=KEY_PATH -n NAMESPACE

    Replace the following:

    • SECRET_NAME: the name of the Kubernetes secret that you are creating—for example, gcs-key.

    • KEY_PATH: the local file system path to the key.json file that you downloaded in the previous steps.

    • NAMESPACE: the namespace of the database cluster.

  7. Apply the following manifest:

      apiVersion: alloydbomni.dbadmin.goog/v1
  kind: BackupPlan
  metadata:
    name: BACKUP_PLAN_NAME
    namespace: NAMESPACE
  spec:
    dbclusterRef: DB_CLUSTER_NAME
    backupSchedules:
      full: "FULL_CRON_SCHEDULE"
      differential: "DIFF_CRON_SCHEDULE"
      incremental: "INCR_CRON_SCHEDULE"
    backupRetainDays: RETENTION_DAYS
    paused: PAUSED_BOOLEAN
    backupLocation:
      type: GCS
      gcsOptions:
        bucket: BUCKET_URL
        key: BACKUP_PATH
        secretRef:
          name: SECRET_NAME
          namespace: NAMESPACE

    Replace the following:

    • BACKUP_PLAN_NAME: a name for this backup plan resource—for example, backup-plan-1.

    • NAMESPACE: the Kubernetes namespace for this backup plan. It must match the namespace of the database cluster.

    • DB_CLUSTER_NAME: the name of your database cluster, which you assigned when you created it.

    • FULL_CRON_SCHEDULE: a backup schedule to create a full backup, containing all data, expressed in cron format. For example, set to "0 0 * * 0" to take a full backup at 00:00 on every Sunday.

    • DIFF_CRON_SCHEDULE: a backup schedule to create backups that are initially full backups. Subsequent backups are differential, based on intervening changes to the data, expressed in cron format. For example, set to "0 22 * * 3"to take a differential backup at 22:00 on every Wednesday.

    • INCR_CRON_SCHEDULE: a backup schedule to create backups that include data that has changed from the last full, differential, or incremental backup. It is expressed in cron format. For example, set to "0 21 * * *" to take an incremental backup at 21:00 every day.

    • RETENTION_DAYS: the number of days for which the AlloyDB Omni Operator retains this backup. It must be an integer between 1 and 90. The default value is 14.

    • PAUSED_BOOLEAN: specifies whether or not the backup plan is paused. Provide one of the following values:

      • true: backups are paused and no scheduled backups are created.

      • false: the AlloyDB Omni Operator creates backups according to the schedule specified by cronSchedule. This is the default value, if not explicitly set to true.

      The default value is false.

    • BUCKET_URL: the name of the Cloud Storage bucket that you created in an earlier step. This is not the full URL to the bucket; don't prefix the bucket name with gs://.

    • BACKUP_PATH: the path of the directory that the AlloyDB Omni operator writes backups into, within the Cloud Storage bucket. The path must be absolute, beginning with /.

    • SECRET_NAME: the name you chose for the Kubernetes secret that you created in an earlier step.

Manually create a backup

You can manually create a backup resource at any time, using any backup plan that you have already applied to a database cluster. The AlloyDB Omni operator applies the chosen backup plan's storage location and retention period to the new manual backup.

To manually create a backup, apply the following manifest:

apiVersion: alloydbomni.dbadmin.goog/v1
kind: Backup
metadata:
  name: BACKUP_NAME
  namespace: NAMESPACE
spec:
  dbclusterRef: DB_CLUSTER_NAME
  backupPlanRef: BACKUP_PLAN_NAME
  manual: true
  physicalBackupSpec:
    backupType: BACKUP_TYPE

Replace the following:

  • BACKUP_NAME: a name for this backup—for example, backup-1.

  • NAMESPACE: the Kubernetes namespace of this restore. It must match the namespace of the database cluster.

  • BACKUP_PLAN_NAME: the name of the backup plan resource that this backup belongs to. It must match the name you chose when you created the backup plan.

  • DB_CLUSTER_NAME: the name of your database cluster, which you assigned when you created it.

  • BACKUP_TYPE: specifies the type of manual backup you want to create. Choose from one of the following values:

    • full: Creates a full backup, containing all data.

    • diff: Creates a differential backup that depends on the last full backup. Subsequent backups are differential, based on intervening changes to the data.

    • incr: Creates an incremental backup that depends on the previous full or differential backup to include data that has changed from the last full or differential.

Monitor and list backups

Your backup plans and backups are all resources on your Kubernetes cluster. To view information about them, use the kubectl get command.

View a backup plan summary

To view information about your database cluster's backup plans, run the following command:

kubectl get backupplan.alloydbomni.dbadmin.goog -n NAMESPACE

Replace NAMESPACE with the namespace of the database cluster.

The output resembles the following:

NAME               PHASE   LASTBACKUPTIME         NEXTBACKUPTIME
backup-plan-prod   Ready   2023-10-26T17:26:43Z   2023-10-27T00:00:00Z

View a list of backups

To view a list of backups available to your database cluster, run the following command:

kubectl get backup.alloydbomni.dbadmin.goog -n NAMESPACE

Replace NAMESPACE with the namespace of the database cluster.

The output resembles the following:

NAME                              PHASE       COMPLETETIME               TYPE
backup-plan-prod-20231026172643   Succeeded   2023-10-26T17:26:53Z       full
manual-backup-1                   Succeeded   2023-10-26T18:15:27Z       full
manual-backup-2                   InProgress                             full

Each row in the output table represents a backup resource, with the following attributes:

  • The name of the backup.
  • The state of the backup, with Succeeded marking a backup ready to restore from.
  • The timestamp of the backup's creation.

Restore from a backup

AlloyDB lets you restore from individual backups or clone a cluster using a backup from a specific point in time.

Restore from a named backup

To restore from a backup, replacing the data in your database cluster with the data in the backup, follow these steps.

  1. List all backups whose phase is Succeeded.

    kubectl get backup.alloydbomni.dbadmin.goog -n NAMESPACE | grep Succeeded

    Replace NAMESPACE with the namespace of the database cluster.

    If at least one good backup candidate exists, then the output resembles the following:

    backup-plan-prod-20231026172643   Succeeded   2023-10-26T17:26:53Z
manual-backup-1                   Succeeded   2023-10-26T18:15:27Z

  2. Choose one of the backups listed in the previous step as the backup to restore from. Take note of its name, which you use in the following step.

  3. Apply the following manifest:

      apiVersion: alloydbomni.dbadmin.goog/v1
  kind: Restore
  metadata:
    name: RESTORE_NAME
    namespace: NAMESPACE
  spec:
    sourceDBCluster: DB_CLUSTER_NAME
    backup: BACKUP_NAME

    Replace the following:

    • RESTORE_NAME: a name to use with the data-restore resource that this manifest creates—for example, restore-1.

    • DB_CLUSTER_NAME: the name of your database cluster, which you assigned when you created it.

    • BACKUP_NAME: the name of the backup that you chose in the previous step.

Clone a cluster from a point in time

AlloyDB Omni operator lets you clone cluster's data from any point in time within a recovery window. The length of the recovery window is directly determined by the retention period.

For example, if your retention period is set to 14 days, then you cannot recover data older than 14 days. You can restore to any point in time within the recovery window. AlloyDB Omni operator retains backups and logs for one day longer than the specified value.

  1. Monitor your recovery window to identify the point of restore:

    kubectl get backupplan.alloydbomni.dbadmin.goog BACKUP_NAME -n NAMESPACE -o json | jq .status.recoveryWindow

    The following shows an example response:

    recoveryWindow:
begin: "2024-01-31T02:54:35Z"

    The timestamp value in the RFC 3339 timestamp format is used in the restore resource.

  2. Create and apply the following restore resource manifest:

    apiVersion: alloydbomni.dbadmin.goog/v1
kind: Restore
metadata:
  name: RESTORE_NAME
  namespace: NAMESPACE
spec:
  sourceDBCluster: DB_CLUSTER_NAME
  pointInTime: "DATE_AND_TIME_STAMP"
  clonedDBClusterConfig:
    dbclusterName: NEW_DB_CLUSTER_NAME

    Replace the following:

    • RESTORE_NAME: a name to use with the data-restore resource that this manifest creates—for example, restore-1.

    • DB_CLUSTER_NAME: the name of your database cluster, which you assigned when you created it.

    • DATE_AND_TIME_STAMP: the RFC 3339 timestamp at minute granularity of the continuous backup you want to restore from—for example, 2024-03-05T15:32:10Z.

    • NEW_DB_CLUSTER_NAME: the name of the new database cluster.

View restore status

  1. View progress of the restore operation:

    kubectl get restore.alloydbomni.dbadmin.goog -n NAMESPACE

    Replace NAMESPACE with the namespace of the database cluster.

    To run the command continuously, add the flag -Aw.

    The output resembles the following:

    NAME        PHASE               COMPLETETIME   RESTOREDPOINTINTIME
restore-1   RestoreInProgress

    When the value of the PHASE column in the output table reads ProvisionSucceeded, then the restore is complete.

  2. View progress of the restored or cloned database cluster coming online:

    kubectl get dbclusters -A -n NAMESPACE

    Replace NAMESPACE with the namespace of the database cluster.

    To run the command continuously, add the flag -Aw.

    The output resembles the following:

    NAMESPACE   NAME               PRIMARYENDPOINT   PRIMARYPHASE   DBCLUSTERPHASE
default     db-cluster-1       10.128.0.55       Ready          DBClusterReady

    When the value of the DBCLUSTERPHASE column in the output table reads DBClusterReady, then the restored or cloned database cluster is ready for use.

Delete a backup

Usually, you don't need to manually delete backups. The AlloyDB Omni operator automatically deletes backups older than the retention period that you specify when creating a backup plan.

If you do want to manually delete a backup, the backup must meet the following requirements:

  • The backup is not the only backup stored for its backup plan. The AlloyDB Omni operator requires that at least one backup exists per backup plan.

  • The backup has no other backups depending on it. For example, a full backup with differential or incremental backups that depend on it; or an incremental backup with differential backups that depend on it.

To delete a backup, run the following command:

kubectl delete backup.alloydbomni.dbadmin.goog/BACKUP_NAME -n NAMESPACE

Replace the following:

  • BACKUP_NAME: the name of the backup to delete.

  • NAMESPACE: the namespace of the database cluster.

Resize a backup disk

To resize the local disk that stores your backups in the Kubernetes cluster, complete the following steps:

  1. Update the resources.disks field of the DBCluster manifest as follows:

    spec:
  primarySpec:
    resources:
      disks:
        - name: BACKUP_DISK
          size: 10Gi

    Replace BACKUP_DISK with the name of the disk storing your backups.

  2. Apply the manifest to enforce the update.

    The AlloyDB Omni operator applies the updated specifications to your DBCluster immediately.

The following restrictions apply to modifying a running database cluster's backup disk:

  • You can increase a disk's size only if the specified storageClass supports volume expansion.
  • You can't decrease a disk's size.

Update a backup plan

Every backup plan is a Kubernetes resource. To update its configuration, do one of the following:

  • Edit and re-apply the manifest file of the backup plan.

  • Use the kubectl patch command.

For example, to pause a running backup plan, change the paused attribute of its manifest to true, and then re-apply the manifest.

Delete a backup plan

To delete a backup plan and remove all of its backup resources, run the following command:

kubectl delete backupplan.alloydbomni.dbadmin.goog/BACKUP_PLAN_NAME -n NAMESPACE

Replace the following:

  • BACKUP_PLAN_NAME: the name of the backup plan to delete.

  • NAMESPACE: the namespace of the database cluster.

To pause a backup plan without deleting it, set the paused attribute of the backup plan's resource to true. A paused backup plan continues to store backups and allow manual backup creation.