A rolling restart is shutting down and updating nodes one at a time (while the other nodes are running) until they’re all updated. This keeps your site running while you update your cluster, whether it’s physical, container, or image based.
Here’s how a rolling restart works:
Shut down one cluster node (JVM instance).
Update/modify the deployment for that node (see the maintenance scenarios that follow).
Start the node.
Repeat these steps for all other cluster nodes.
User experience can be inconsistent during a rolling restart. For example, UI changes in a plugin update are only visible on the updated nodes. Users on nodes that haven’t been updated see the old interface. Maintenance scenarios might have specific cases that cannot be performed in rolling restarts. The scenario descriptions mention these cases.
The maintenance scenarios eligible for rolling restarts are described below.
For a new plugin or module (one that does not already exist in the cluster) to be eligible for rolling restart it must not modify data, delete, or rename database columns in a way that breaks compatibility with existing plugins or modules.
For a new version of an existing plugin or module to be eligible for rolling restart, it must not modify data or delete or rename database columns in a way that breaks compatibility with the existing version of the plugin or module.
The Customer Portal identifies fix packs that are not revertible, and therefore ineligible for rolling restart. All other fix packs are eligible.
Revertible fix packs can be removed in rolling restarts.
Portal Properties file changes can be applied in rolling restarts.
System configuration files can be applied in rolling restarts.
Modifications to application server and JVM settings can be done in rolling restarts.
Minor version updates of Java can be applied in rolling restarts. Major version updates are not supported in rolling restarts, and should instead be done when all cluster nodes are shut down.
All rolling restart-eligible updates can be applied using the rolling restart steps listed above. Other updates must be done differently as described next.