We have cluster of 2 nodes, and observe huge delay when jms starts, particularly when we have the great number of messages e.g. 100 000 000 in queues. After crash in table JBM_MSG_REF remain all informations about relations msg to nodes. First what server does when it starts is update this references, but not in a one bulk update, but updating one by one every row.
When in JBM_MSG_REF is the vast number of tuples that increase a number of updates need to be done on startup, and it longer significantly time to stand up after crash. Sometimes it takes our servers 20 minutes to start after fail.
Is any way to speed up the failover startup?
We use JBoss 5.1.GA, JBoss Messaging 1.4.8 and Oracle as a persistence for jms. The same efect we observe using any earlier version 1.4.x.