I'm using Solr 4.7.2. A few things I've missed follow. Before reaching the "one leader-one failed to recover" state, the situation was no leader for the shard and both nodes in "recovery failed" mode. A bit of tinkering to clusterstate.json "forced" the one to be the leader but that didn't change a thing. The error message "No registered leader was found after waiting" was the same before and after tinkering.
The other thing I found out is that there was a period when ZK ensemble was down for a while. In Solr log and upon startup I found ithe following org.apache.zookeeper.KeeperException$NoNodeException: KeeperErrorCode = NoNode for /collections/sample/leaders/shard1 It seems though that the Solr servers could not communicate with ZK once it was functional again. Restarting the servers once more fixed the issue and the failing server managed to "recover" the index successfully. -- View this message in context: http://lucene.472066.n3.nabble.com/SolrCloud-node-recovery-fails-with-No-registered-leader-was-found-tp4137331p4137584.html Sent from the Solr - User mailing list archive at Nabble.com.
