You will want to start zookeeper independently of Solr. If you want fault tolerance with regard to Zookeeper, you should have three of them. Hosting at Amazon, for example, I'd have one Zookeeper in each of three availability zones, meaning that any one AZ can go down, leaving me with two functioning zookeeper nodes, which is sufficient for my Solr servers to continue operating. (Zookeeper requires half+1 of your servers to be up to function properly, thus an odd number of servers).
Upayavira On Fri, Mar 1, 2013, at 11:41 AM, AlexS wrote: > Hello, I'm new to Solr, and I need to create the smallest possible Solr > cluster, supporting replication and fault tolerance, with no single point > of > failure. > I've followed the instructions at http://wiki.apache.org/solr/SolrCloud, > and > that's fine, but since my dataset it's not that large, I don't really > need > sharding the index. > So what I would need, it's having a whole index in one host, a whole > replica > in a different host, and I'll use a load balancer to hit any of the boxes > for searching and indexing. > I've tried this configuration, with one ZK starting in host1. Seems to > work > fine if I stop host2, but if I stop host1 (the one containing ZK), host2 > start throwing java.net.ConnectException, and even though searching is > still > possible, indexing fails with error 503. > I've also tried to start a ZK in both hosts, but this fails to start, it > looks like ZK instances need to be in an odd number, because of the > voting > mechanism? > > So what are my options here? Am I doing something wrong with Solr/ZK? > > Thanks in advance for your help. > > > > -- > View this message in context: > http://lucene.472066.n3.nabble.com/Minimum-Solr-cluster-configuration-tp4043943.html > Sent from the Solr - User mailing list archive at Nabble.com.
