Backing out a rolling upgrade

Important Caveat

It is possible to back out configuration changes on a given server application instance, depending on the state of the configuration deployment on that instance. Guidewire does not recommend that you back out an ongoing deployment on a server instance except in extraordinary circumstances. If you do so, you need to test that the server configuration returns to a safe state.

Backing out schema changes

To support backing out a rolling upgrade, Guidewire prevents a user from writing to any table column newly added during the upgrade process until the rolling upgrade is fully complete:
  • During the rolling upgrade, PolicyCenter tracks which table columns the upgrade process adds and stores the list of new columns in the database.
  • For all cluster servers, PolicyCenter reads the list on server start up and throws an exception for that server if a user tries to write to a newly added column before the rolling upgrade is complete.
  • It is only after the rolling upgrade completes (a user clicks Rolling Upgrade Complete in the Server Tools Upgrade and Versions screen) that PolicyCenter permits write operations to any table column added during the rolling upgrade.

Enforcing this restriction makes rolling back data model schema changes easier as the rollback process does not have to contend with data inconsistencies between the different cluster nodes.

The back out process does not undo any schema changes made during the rolling upgrade process other than changes to typelists.

Initiating the back out of a rolling upgrade

Important: Before you begin the process of backing out a rolling upgrade of a PolicyCenter, first shut down any cluster node that is running the new, target, application configuration.

If a rolling upgrade of a PolicyCenter cluster is in progress, the PolicyCenter Server Tools Upgrade and Versions screen shows an Initiate Backout button. Clicking this button opens the back-out wizard screen. This screen provides on-screen instructions on how to proceed with the back out of the rolling upgrade.

As part of the back out process, PolicyCenter runs an internal batch process to verify there are no orphaned typecodes. At least one cluster member must have the batch server role in order to execute this process.

Rolling Upgrade Failure

Under rare circumstances, it is possible for the failure of a rolling upgrade on a specific application server to result in the inability to back out of the rolling upgrade on that server. To provide for this upgrade scenario, Guidewire provides a Force Backout button on the Upgrade and Versions screen. Clicking this button brings up a dialog in which you can select the server ID of the inactive server. The Force Backout button is only available if a rolling upgrade is in progress.