On 7/27/2015 10:59 PM, Modassar Ather wrote: > If we upgrade zookeeper we need to restart. This upgrade process is > automated for future releases/changes of zookeeper. > This is a single external zookeeper which is completely stopped/shutdown. > No Solr node are restarted/shutdown. > What I have understanding that even if the zookeeper shuts down, after > restart the Solr nodes should come insync with the ZK state. Please correct > me if I am wrong.
Disclaimer: I do not have a ton of concrete experience with SolrCloud. I do have a cloud setup, but it is running Solr 4.2.1, which at this point is ancient. I haven't needed to do much to maintain it ... it takes care of itself. Recovering correctly from a complete zookeeper failure is what I would hope for, but it's a scenario that I've never tried. I hope there's a unit test for it, but I haven't checked. A fully redundant zookeeper ensemble requires a minimum of three hosts. If you need to upgrade ZK, then you upgrade them one at a time, and the ensemble never loses quorum. http://wiki.apache.org/hadoop/ZooKeeper/FAQ#A6 Thanks, Shawn
