Problem Each cluster member dispatches to the users (notification menu, mails) only the events it produced itself, and since XWIKI-25258 it also resumes only its own pending live notification mails after a restart. When a member never comes back with the same identifier (scale down of the cluster, identifier changed), the events it had not dispatched yet are never shown to the users, by alert or by mail, and its pending live notification mails are never sent. Found while testing XWiki on Kubernetes, where scaling down the cluster removes a member for good. Proposal Once a member has been absent for observation.remote.reclaimDelay, the cluster singleton node:
- sends its pending live notification mails, for the events produced before it left;
- dispatches its events which were not dispatched yet, so they appear in the notification menu and their mails are sent.
A member which comes back after that finds nothing left to do for these events. Remote observation disabled: no change. |
