On 2/5/2016 8:56 AM, Keith L wrote: > Caches are stored on the Java heap for each instance of a searcher. The > filter cache would be different per replica, same for the doc cache, and > query cache > > On Fri, Feb 5, 2016 at 8:47 AM Tom Evans <tevans...@googlemail.com> wrote: > >> I have a small question about fq in cloud mode that I couldn't find an >> explanation for in confluence. If I specify a query with an fq, where >> is that cached, is it just on the nodes/replicas that process that >> specific query, or will it exist on all replicas?
Keith already answered this. Cache entries will only exist on the node where the query was handled. >> We have a sub type of queries that specify an expensive join condition >> that we specify in the fq, so that subsequent requests with the same >> fq won't have to do the same expensive query, and was wondering >> whether we needed to ensure that the query goes to the same node when >> we move to cloud. SolrCloud will normally load balance queries across the entire cluster, even if you send every query to the same node. This is inherent in the design of SolrCloud. That behavior can be overridden with the "preferLocalShards" option. This option *might* help performance in a very lightly used cluster, but is NOT recommended for a busy cluster. See the end of this page (before the comments): https://cwiki.apache.org/confluence/display/solr/Distributed+Requests Here's the issue where the option was added to version 5.1: https://issues.apache.org/jira/browse/SOLR-6832 Thanks, Shawn
