With HA cluster, when there are 1600 paging directories, it could be very slow to finish failover. here is the summary:

1. With 3200 addresses, it spent 15 seconds on failover.
2. Producing 1600 paging dir under data/paging, failover takes 60+ seconds.

This happened when broker1 failovers to broker2 in a HA cluster with a shared NFS storage.

We are helping one big client to migrate to AMQ, this delay on failover could hinder the decision, I suggest we could prioritize this task and make a implementation plan for it.

To fix that, we added purge-page-folders on broker.xml, or purgePageFolders (with default = false) on broker properties.