| Upgrading a cluster safely needs the database migrations and the Distribution Wizard to run once, before the members of the new version serve requests, and the upgrade process needs to know whether it worked (for example a Kubernetes Job run before a rolling update). The existing xwiki.store.migration.exitAfterEnd property of xwiki.cfg does not allow it: it stops the JVM with the exit code 0 in the middle of the initialization, before the Distribution Wizard (so pages and extensions are not upgraded, and an empty database is left without any page), and on a failed migration the JVM keeps running.
Proposal: a core.initialization.exitAfterUpgrade property of xwiki.properties (implies the eager initialization of XWIKI-25247) which waits for the initialization, checks the database of every wiki, runs the Distribution Wizard (not interactive) on the main wiki then on every sub-wiki, checks the results (wizard errors, failed extension jobs, installed user interface version), logs a summary and stops the JVM with the exit code 0 (upgraded), 1 (initialization or database failure) or 2 (Distribution Wizard failure). xwiki.store.migration.exitAfterEnd becomes deprecated. |
