I’m not a fan of CloudSolrClient. I think it adds failure modes without any meaningful speedup. It adds an unnecessary dependency on the Zookeeper ensemble. A request sent to any node can be handled correctly without bothering Zookeeper. Every node already has full information about the cluster topology.
Put a load balancer in front of your cluster and use that. I’ve been able to load very large clusters (48 hosts, 2000+ CPUs) to max with simple load balancers (AWS ELB or ALB). A Solr Cloud cluster with a down Zookeepers ensemble should work normally until there is a topology change, like a replica going down. Pretty easy to test, shut down all the Zookeeper nodes in the middle of a load test. wunder Walter Underwood wun...@wunderwood.org http://observer.wunderwood.org/ (my blog) > On May 21, 2024, at 8:30 AM, Matt Kuiper <kuipe...@gmail.com> wrote: > > Thanks for the responses! > > Shawn, > > I believe I understood that Zookeeper does not actually handle the request. > Thanks for the clarification as I did not word it properly. > > More precisely, will a SolrJ CloudSolrClient work for the purpose of > querying SolrCloud if the ZK host list it is configured with represents a > Zookeeper ensemble that has lost its quorum? > > My goal is to learn if this SolrJ client (configured via zkhost list) can > reliably query a SolrCloud even when the ZK ensemble has lost its quorum. > > I have heard that Zookeeper will go into a read-only mode if quorum is > lost. Put another way, will the SolrJ cloud client still be able to talk > to ZK to discover the URLs required to get > to Solr and serve queries (at least will the solrj client be given the > latest state from the remaining ZK nodes for what was last written to > Zookeeper)? > > Our company is reviewing HA/DR scenarios for SolrCloud and I am > investigating if a SolrCloud that has lost its ZK quorum can provide any > value for Solr queries, and how to characterize accuracy of any search > results. > > Thanks for any further information you can provide. > > Matt > > On Mon, May 20, 2024 at 9:02 PM Shawn Heisey <apa...@elyograg.org.invalid> > wrote: > >> On 5/20/2024 17:46, mtn search wrote: >>> Will a SolrJ client connection (configured with a ZKhost list) to a >>> SolrCloud still work for querying a SolrCloud when the Zookeeper ensemble >>> has lost its quorum, but some ZK nodes remain up? I understand that >>> indexing will not work. >>> >>> Follow-on - will the Solr query work to the same SolrCloud if the query >> is >>> sent directly to a Solr node rather than to Zookeeper? >> >> Requests to SolrCloud such as queries or indexing do not go through ZK, >> ever. >> >> The SolrJ cloud client talks to ZK to discover the URLs required to get >> to Solr and which of them are online at any given moment. ZK is used as >> a central database to coordinate the cluster, it does not handle requests. >> >> Thanks, >> Shawn >> >>
