With a remote Solr server, the cluster singleton node runs a single synchronization job for the whole farm, whatever solr.synchronizeAtStartupMode says (one job per wiki would run all the wiki synchronizations at the same time). The WIKI mode keeps its meaning with the embedded Solr server.
A member which becomes the singleton node waits solr.synchronizeOnSingletonGainDelay seconds (new property, 60 by default) before synchronizing, and does nothing if it is not the singleton node anymore by then. A short restart of the preferred member (for example a member dedicated to background tasks with a higher observation.remote.singletonPriority) therefore does not trigger a synchronization on another member. When the preferred member is down longer than the delay, it synchronizes again when it comes back, so the index is fully compared twice in that case.
The behaviour is documented in xwiki.properties next to solr.synchronizeAtStartup and solr.synchronizeAtStartupMode.
This message was sent by Atlassian Jira (v9.3.0#930000-sha1:287aeb6)
If image attachments aren't displayed, see this article.
