Several update scenarios cannot be done by rolling restart because they affect cluster communication, break compatibility with existing plugins/modules, or break Service Builder services. Also non-revertible updates are disqualified because reversing their effects requires restoring data from a backup.
Maintenance changes ineligible for rolling restart are typically done on all nodes at once, when they’re shut down. The following sections describe techniques for applying these changes.
- Custom Plugin/Module Data Schema Changes
- Custom Plugin/Module Data Changes
- Non-revertible Fix Packs (DXP only)
- Service Builder Service Schema Version Changes
- Cluster Code Changes
Data schema changes are explained first.
Custom Module/Plugin Data Schema Changes
Custom module/plugin data schema changes that break compatibility with existing modules and plugins must be introduced over several releases in which the data is transitioned and maintained in old and new columns until the old column is no longer needed.
Custom Module/Plugin Data Changes
Data changes to modules or plugins you’ve developed require these steps:
Create a new column.
Copy the data to the new column.
Maintain both columns until the old column is no longer used by any cluster nodes.
Delete the column in the next release.
Non-revertible Fix Packs (DXP only)
The Customer Portal identifies fix packs that are not revertible. Non-revertible fix packs must be applied to nodes when they are all shut down.
Service Builder Service Schema Version Changes
Liferay-Require-SchemaVersion (specified in its
match the module’s schema version value in the
Release_ table. Installing a
new module that has a new schema version updates the
Release_ table with that
schema version and triggers a data upgrade process. If you install such a module
on one node, the schema version in the
Release_ table no longer matches the
Liferay-Require-SchemaVersion of the modules on the other nodes. The module’s
Service Builder services become unavailable. Such changes cannot be reverted:
the database must be restored from a backup. These schema version changes must
be applied while all nodes are shut down.
Cluster Code Changes
Cluster communication must stay intact. For this reason, cluster code must not be updated in rolling restarts. The Customer Portal identifies DXP fix packs that contain such changes as non-revertible. Here are packages you must not change in rolling restarts:
Now you know how to update your cluster using ways other than rolling restart.