Replicated LevelDB cluster: failed to page in queue messages

Hi,

I faced following exception while testing ActiveMQ leveldb cluster.
On this exception, master demotes itself to slave. New master faces same
exception, and it keeps on happening. Master keeps on rotating without
processing messages.
Client was not able to send/receive messages.

I've opened jira issue: https://issues.apache.org/jira/browse/AMQ-5459

Any idea what could cause this - leveldb corrupted?
Is there any way to recover out of this without loosing existing messages in
queues?
I would highly appreciate any pointers to fix this.

Restarting activemq on all cluster nodes did not help, it ran into same
issue.

Nov 25 15:03:35.777 [esta] Scheduler] - ERROR AbstractStoreCursor - cbtesta

org.apache.activemq.broker.region.cursors.QueueStorePrefetch [ at ] 72e4521c:testpt,batchResetNeeded=false,storeHasMessages=true,size=2516,cacheEnabled=false,maxBatchSize:200,hasSpace:true
- Failed to fill batch
Nov 25 15:03:35.777 [esta] Scheduler] - ERROR AbstractStoreCursor - cbtesta

org.apache.activemq.broker.region.cursors.QueueStorePrefetch [ at ] 72e4521c:testpt,batchResetNeeded=false,storeHasMessages=true,size=2516,cacheEnabled=false,maxBatchSize:200,hasSpace:true
- Failed to fill batch