Reported on the user list and evident in tests.
So far this is only a known issue for 4.x
In 5.x I am already using core.getNewIndexDir() when getting the file list for replication - I must have changed to that doing the Directory based replication work.
4.x is still using core.getIndexDir().
[trunk commit] Mark Robert Miller
SOLR-4168: 5x changes entry
[branch_4x commit] Mark Robert Miller
SOLR-4168: Ensure we are using the absolute latest index dir when getting list of files for replication.
That should solve a couple of the recent 4x test failures.
So something is still happening here - doesn't seem that was the issue or only issue. Still not a problem seen on 5.x though.
This was causing most of the fails waiting on recovery in 4.x tests. I still don't know why it was happening, but it has not shown up with the now back ported 5x replication code that uses a Directory.