Skip to main content
Version: 2.12

Disaster Recovery Checkpoint

Disaster Recovery is an essential part of a High Availability setup.

In a disaster recovery scenario where there has been a full Kubernetes failure, the goal is to relaunch Ververica Platform with all deployments starting from the latest known state. To enable this, Ververica Platform stores a permanent reference in the job to a Disaster Recovery Checkpoint.

Configuration​

The elapsed time between Disaster Recovery Checkpoints is determined by the checkpoint configuration you set and the Controller monitoring frequency for Ververica Platform. The monitoring interval is 3s.

To configure a Disaster Recovery Checkpoint, set a value n for the disasterCheckpointsDelay configuration property in your Deployment configuration YAML file. With the property enabled, every n-th standard Flink checkpoint will be saved to the job as the updated Disaster Recovery Checkpoint.

Additionally, set state.checkpoints.num-retained for your Deployment via the UI as Additional Deployment configuration to save the last m standard Flink checkpoints to Blob Storage. As a rule of thumb choose a number equal to the disaster checkpoints delay value plus 1. The default value is 1:

Set the Checkpoints Value

Set the disasterCheckpointsDelay configuration property in your Deployment configuration YAML file:

     metadata:
       displayName: disaster-checkpoints
     spec:
       deploymentTargetName: vvp-jobs
       template:
         spec:
           artifact:
             jarUri: >-
               s3://vvp-snapshot-blob-storage-eu-west-1/artifacts/namespaces/default/TopSpeedWindowing.jar
             kind: JAR
           disasterCheckpointsDelay: 5

where:

  • A value of 0 disables saving the Disaster Recovery Checkpoint.
  • An integer value of > 0 specifies which Flink checkpoint will be saved in the job as the Disaster Recovery checkpoint, so that for a value n, every n-th checkpoint is saved.

Ensure that you set the required Additional Configuration for the Deployment:

  • state.checkpoints.num-retained: m, where m is typically disasterCheckpointsDelay value +1.

Ververica Platform checks the value during Deployment validation. If no value is set, or if the value set is too low, Ververica Platform shows a pop-up warning and suggests a value:

No Checkpoints Configured

For more about configuring a Deployment see the Deployments documentation, which includes a full configuration example.