Hi Eric, Sorry I did not reply earlier. I see this page cached here - on gmane.org but the original post I posted on Solr Users list does not show your comment - http://lucene.472066.n3.nabble.com/Leaders-in-Recovery-Failed-state-td4180610.html
I'm on Solr 4.10.1 - The last time this had happened I removed the replica for the affected Shards (the shards where the Leaders were shown as Down) - deleted the Replica data directories and then added the replica back using the Collections API - the did the trick then (but I'n not sure if that was the right way to do it). Also the problem seemed to have rooted from the fact that the Zookeeper instances were on the same machines as the Solr servlet containers and perhaps the Zookeeper instances were starved of resource (CPU & disk) - I have had since moved the Zookeeper instances out to separate servers and that makes the boot time fast - but not all shards come online when all the solr cloud instances are reboot. A few servers from the Solr Cluster went down again and I have the same issues where for 3 shards the Leaders are shown as down and the logs in the log files for these instances as below - INFO - 2015-02-09 05:18:13.696; org.apache.solr.handler. admin.CoreAdminHandler; In WaitForState(recovering): collection=collection1, shard=shard10, thisCore=collection1_shard10_replica2, leaderDoesNotNeedRecovery=false, isLeader? true, live=true, checkLive=true, currentState=recovering, localState=down, nodeName=10.68.77.8:8983_solr, coreNodeName=core_node28, onlyIfActiveCheckResult=true, nodeProps: core_node28: {"state":"recovering","core":"collection1_shard10_replica1", "node_name":"10.68.77.8:89 83_solr","base_url":"http://10.68.77.8:8983/solr"} I have tried deleting the replica for these shards - but this time the Delete Replica Async requests are showing in "submitted" state for very long now (over 3 hours) - last time when I did this these requests finished fairly quickly. Any pointers are greatly appreciated. Thanks, Anand