Roll back application upgrades

If you have issues with your application or find inconsistencies in your application code between Flink versions, you can roll back using the AWS CLI, AWS CloudFormation, AWS SDK, or the AWS Management Console. The following examples show what rolling back looks like in different failure scenarios.

Runtime upgrade succeeded, the application is in RUNNING state, but the job is failing and continuously restarting

Assume you are trying to upgrade a stateful application named TestApplication from Flink 1.15 to Flink 1.18 in US East (N. Virginia). However, the upgraded Flink 1.18 application is failing to start or is constantly restarting, even though the application is in RUNNING state. This is a common failure scenario. To avoid further downtime, we recommend that you roll back your application immediately to the previous running version (Flink 1.15), and diagnose the issue later.

To roll back the application to the previous running version, use the rollback-application AWS CLI command or the RollbackApplication API action. This API action rolls back the changes you've made that resulted in the latest version. Then it restarts your application using the latest successful snapshot.

In enhanced snapshot mode, Managed Service for Apache Flink restores the newest READY enhanced snapshot at or before the application version that the rollback returns to. For more information, see Roll back an application.

We strongly recommend that you take a snapshot with your existing app before you attempt to upgrade. This will help to avoid data loss or having to reprocess data.

In this failure scenario, CloudFormation will not roll back the application for you. You must update the CloudFormation template to point to the previous runtime and to the previous code to force CloudFormation to update the application. Otherwise, CloudFormation assumes that your application has been updated when it transitions to the RUNNING state.

Rolling back an application that is stuck in UPDATING

If your application gets stuck in the UPDATING or AUTOSCALING state after an upgrade attempt, Amazon Managed Service for Apache Flink offers the rollback-applications AWS CLI command, or the RollbackApplications API action that can roll back the application to the version before the stuck UPDATING or AUTOSCALING state. This API rolls back the changes that you’ve made that caused the application to get stuck in UPDATING or AUTOSCALING transitive state.

In enhanced snapshot mode, RollbackApplication restores from an enhanced snapshot. For an application in the UPDATING or AUTOSCALING status, Managed Service for Apache Flink uses the newest READY enhanced snapshot of the application. If the application has no READY enhanced snapshot, the rollback returns InvalidRequestException.

The IAM principal that calls RollbackApplication needs the kinesisanalytics:RestoreFromSnapshot permission on the selected enhanced snapshot. For more information, see Roll back an application.

Roll back after an upgrade to enhanced snapshot mode

After a rollback to the application version before the upgrade to enhanced snapshot mode, the application uses application snapshots (legacy) again. Managed Service for Apache Flink restores the newest READY application snapshot (legacy) at or below the application version before the upgrade.

• The rollback works only if you did not update the application after the upgrade to enhanced snapshot mode.

• If no READY application snapshot (legacy) exists at or below the application version before the upgrade, RollbackApplication returns InvalidRequestException.

• After the rollback, the application uses application snapshots (legacy). The application can use the enhanced snapshots that it created after the upgrade only after you upgrade it to enhanced snapshot mode again.

For more information about rollbacks across the boundary of the upgrade to enhanced snapshot mode, see Roll back after the upgrade.