Inconsistent numDocs between leader and replica
As of 4.6.1, a work around is to be sure to call blockUntilFinished, which will spawn a new Runner thread if needed to make sure the queue is emptied.
I think that this will address it. I've been testing it with a case that I found while working on SOLR-4260.
Commit 1559620 from Mark Miller in branch 'dev/trunk'
[ https://svn.apache.org/r1559620 ]
SOLR-5643: ConcurrentUpdateSolrServer will sometimes not spawn a new Runner thread even though there are updates in the queue.
Commit 1559621 from Mark Miller in branch 'dev/branches/branch_4x'
[ https://svn.apache.org/r1559621 ]