Major Description
The problem we saw was that when the solr overseer leader experienced temporary zk connectivity problems it stopped processing overseer queue events.
This first happened when quorum within the external zk ensemble was lost due to too many zookeepers being stopped (similar to SOLR-5199). The second time it happened when there was a sufficient number of zookeepers but they were holding zookeeper leadership elections and thus refused connections (the elections were taking several seconds, we were using the default zookeeper.cnxTimeout=5s value and it was hit for one ensemble member).