Just pointing out the obvious, any node can answer to any query by relaying it to the node that actually hosts a replica of a shard of the collection(s) being queried. It’s just an extra hop if the asked node doesn’t hold a replica
-Ufuk — > On Apr 2, 2025, at 20:59, Jan Høydahl <jan....@cominvent.com> wrote: > > Hi, > > Any of the above :) > > If you don't care about minimizing network hops during indexing/search, you > can use a LB like HAProxy, NginX or some hardware product in front of your > nodes for simplified logic in your application. > > However, the best practice these days is to provide your application with a > list of Solr host names and initialize a Solr Client (e.g. for Java the > CloudSolrClient) with several host names. The client will then reach out to > one of these hosts and request the cluster state which lists all nodes, which > of them are alive etc. And then the Solr Client will manage routing for you > directly to the best node, and also fail over. If your programming language > does not have a SolrClient that can do this then implement this in your own > code or consider a LB. > > It is no longer recommended for applications to connect to Zookeeper directly > to get the cluster state. Zookeeper should be protected and reachable only > from the Solr nodes. > > Jan > >> 2. apr. 2025 kl. 11:38 skrev Andreas Mock <andreas.m...@drumedar.de>: >> >> Hi all, >> >> probably for all out there a stupid question, but I didn't find any hint in >> the documentation. >> >> When I install a SolrCloud-Cluster with several nodes, how do you connect to >> any of them to do the job? >> Do you install some kind of proxy in front of the nodes? Or do you ask >> zookeeper for an endpoint of a known node which is up? >> Do you try to connect round robin (a timeout on a not-up node could last too >> long)? >> >> What is the best practice for SolrCloud? >> >> Best regards >> Andreas >> >> >
