I may be interpreting this incorrectly, but shouldn't the cloud still serve requests if ZK crashes?
http://wiki.apache.org/solr/SolrCloud#Example_C:_Two_shard_cluster_with_shard_replicas_and_zookeeper_ensemble " The problem with example B is that while there are enough Solr servers to survive any one of them crashing, there is only one zookeeper server that contains the state of the cluster. If that zookeeper server crashes, distributed queries will still work since the solr servers remember the state of the cluster last reported by zookeeper. The problem is that no new servers or clients will be able to discover the cluster state, and no changes to the cluster state will be possible." Jeremy D. Branham Performance Technologist II Sprint University Performance Support Fort Worth, TX | Tel: **DOTNET Office: +1 (972) 405-2970 | Mobile: +1 (817) 791-1627 http://JeremyBranham.Wordpress.com http://www.linkedin.com/in/jeremybranham -----Original Message----- From: Shawn Heisey [mailto:s...@elyograg.org] Sent: Friday, September 13, 2013 2:40 PM To: solr-user@lucene.apache.org Subject: Re: Best configuration for 2 servers On 9/13/2013 12:50 PM, Branham, Jeremy [HR] wrote: > Does this sound appropriate then? [assuming no 3rd server] > > Server A: > Zoo Keeper > SOLR with 1 shard > > Server B: > SOLR with ZK Host parameter set to Server A Yes, that will work, but if the ZK on server A goes down, the entire cloud is down. When you create a collection with replicationFactor=2, one replica will be on server A and one replica will be on server B. If you want to break the index up into multiple shards, you can, you'll also need the maxShardsPerNode parameter when you create the collection, and all shards will have replicas on both machines. A note about zookeeper and redundancy, and an explanation about why 3 hosts are required: To form a quorum, zookeeper must have the votes of a majority of the hosts in the ensemble. If there are only two hosts, it's not possible for there to be a majority unless both hosts are up, so two hosts is actually worse than one. You need to either have one ZK node or at least three, preferably an odd number. Thanks, Shawn ________________________________ This e-mail may contain Sprint proprietary information intended for the sole use of the recipient(s). Any use by others is prohibited. If you are not the intended recipient, please contact the sender and delete all copies of the message.