Pause a point-in-time recovery

To debug a point-in-time recovery, pause it after the base backup is restored and before the binary logs are applied. Use this only for debugging, not for regular restores. If the recovery fails, see Point-in-time recovery in Troubleshoot backups and restores.

The restore Job creates the sleep-forever file on the restored data directory and waits until you remove it.

1. Add the percona.com/pitr-sleep-forever annotation to the PerconaServerMySQLRestore object. The Operator checks only if the annotation exists, not its value:

   apiVersion: ps.percona.com/v1
   kind: PerconaServerMySQLRestore
   metadata:
     name: restore-pitr
     annotations:
       percona.com/pitr-sleep-forever: "true"
   spec:
     clusterName: ps-cluster1
     backupName: backup1
     pitr:
       type: date
       date: "2026-03-20 09:15:00"
   

2. Apply the object. The point-in-time recovery Job starts a temporary mysqld on the restored data directory, creates the /var/lib/mysql/sleep-forever file, and waits.

3. Find the Pod of the restore Job. The Job is named pitr-restore-<restore-name>. Exec into the Pod to inspect the restored data:

   kubectl get pods -n <namespace> -l job-name=pitr-restore-<restore-name>
   kubectl exec -it <restore-pod-name> -n <namespace> -- bash
   

4. Remove the file when you are done:

   kubectl exec <restore-pod-name> -n <namespace> -- rm /var/lib/mysql/sleep-forever
   

Warning

When you remove the file, the container exits without applying the binary logs. The restore still reports Succeeded, and the Operator starts the cluster with only the data from the base backup, not the data up to your recovery target. Keep your applications away from the cluster, and run the point-in-time recovery again without the annotation. The new restore overwrites any changes made in the meantime.

Warning

The restore Job uses the data volume of the <cluster-name>-mysql-0 Pod, so the sleep-forever file is created on this volume. Always end the pause by removing the file. If you delete the restore object or the Job instead, remove /var/lib/mysql/sleep-forever from the <cluster-name>-mysql-0 Pod afterwards. Otherwise this Pod doesn’t start mysqld.

---

Last update: October 27, 2026
Created: October 27, 2026