[ https://issues.jboss.org/browse/TEIID-2578?page=com.atlassian.jira.plugin... ] Steven Hawkins commented on TEIID-2578: --------------------------------------- If the master is up and later a non-master comes up, that is the issue. The only way this metadata replication works currently is if everyone is up together or if the master joins later. A non-master joining later won't see the affect of the databasestore load - it will only become aware of vdbs as they are further modified and a reloadDatabase event is issued. For example in the internal materialization logic there is handling for the state transfer that happens when someone joins the cluster. Something similar is needed here. Since the master would be issuing a reloadDatabase event, that would replace all of the state on the non-masters. But it does mean that all existing connections would be terminated. There is now only one event distributor method that causes a vdb restart, which is reloadDatabase. Having a more granular versioning could prevent an unnecessary restart of a vdb from the scenario above but that should be a rare occurrence. -- This message was sent by Atlassian JIRA (v6.4.11#64026)