On 6/16/2017 9:05 AM, Xie, Sean wrote: > Is there a way to keep SOLR alive when zookeeper instances (3 instance > ensemble) are rolling updated one at a time? It seems SOLR cluster use > one of the zookeeper instance and when the communication is broken in > between, it won’t be able to reconnect to another zookeeper instance > and keep itself alive. A service restart is need in this situation. > Any way to keep the service alive all the time?
Have you informed Solr about all three of the ZK hosts? You need a zkHost like this, with an optional chroot: zkHost="server1.example.com:2181,server2.example.com:2181,server3.example.com:2181/chroot" There are more zkHost examples and a better description of the string format in this javadoc: https://lucene.apache.org/solr/6_6_0/solr-solrj/org/apache/solr/client/solrj/impl/CloudSolrClient.html#CloudSolrClient-java.lang.String- If Solr hasn't been explicitly informed about all the hosts in the ensemble, then it cannot connect to surviving hosts. I've never heard of the problem you described happening as long as Zookeeper quorum is maintained and Solr is properly configured. If you can show that a correctly configured Solr 6.6 server loses connection to ZK when one of the ZK servers is taken down, that's a bug, and we need an issue in Jira with documentation of the problem. Thanks, Shawn
