Have you deployed ZooKeeper on servers other than the Solr nodes? If you have them on the Solr nodes, then you may be getting elections when under pressure.
-----Original Message----- From: Shawn Heisey [mailto:s...@elyograg.org] Sent: 17 July 2014 14:25 To: solr-user@lucene.apache.org Subject: Re: SolrCloud Issues On 7/17/2014 6:57 AM, Cool Techi wrote: > We have moved to SolrCloud (version 4.8) and are facing several issues > in this setup compared to the master/slave setup we have had for a > while now, > 1) We have a 2 shard set-up with one replica each, we notice that most of the > time the replica's are in recovering status. What could be causing this? > 2) In-consistant results count, especially when one of the nodes is > recovering, I have asked another question earlier on this. To our > understanding the recovering node, doesn't returns any results, what else can > cause this? > 3) Solr node goes down very frequently, there is no OOM or otherwise in the > logs, but a node keeps going down. Also, at times we have noticed that tomcat > stops responding. Since there are so many parts to solr cloud, checking > what's causing the issue is difficult in a quick time, so if anyone else has > faced this, it would be very helpful. I don't have anything specific for you, but if you are having any kind of performance issues at all, it can lead to bizarre SolrCloud behavior. The basic zookeeper client timeout defaults to 15 seconds. This is a very long timeout, but if anything is happening that makes any part of SolrCloud wait longer than 15 seconds, SolrCloud will think there's a problem that needs recovery. Here's a summary of common performance problems and some possible solutions: http://wiki.apache.org/solr/SolrPerformanceProblems Thanks, Shawn
