When combining a load balancer with SolrCloud, the handler definitions in solrconfig.xml should set preferLocalShards to true (which Tom mentioned)
Thanks Shawn! I was wondering where to set this... Yup - my IT guy is sharp, sharp, sharp -- nice to get this confirmation from the list... On Tue, Apr 19, 2016 at 7:59 AM, Shawn Heisey <apa...@elyograg.org> wrote: > On 4/18/2016 11:22 AM, John Bickerstaff wrote: > > So - my IT guy makes the case that we don't really need Zookeeper / Solr > > Cloud... > <snip> > > I'm biased in terms of using the most recent functionality, but I'm aware > > that bias is not necessarily based on facts and want to do my due > > diligence... > > > > Aside from the obvious benefits of spreading work across nodes (which may > > not be a big deal in our application and which my IT guy proposes is more > > transparently handled with a load balancer he understands) are there any > > other considerations that would drive a choice for Solr Cloud (zookeeper > > etc)? > > Erick has a point. If your IT guy feels comfortable with a load > balancer, he should go ahead and set that up. > > For a new install like you're describing, I would probably still use > SolrCloud on the back end, even with a load balancer. > > As Daniel said, a non-cloud replicated setup requires configuration of > masters and slaves. Instead of replication, you could go with a build > system that sends updates to each copy of the index independently. > > When using replication, switching master/slave roles in the event of a > master server failure is not trivial. SolrCloud handles all that, > making multi-server management a LOT easier. Initial setup is slightly > more complicated due to zookeeper, and configuration management requires > an "upload to zookeeper" step ... but I do not think these are not high > hurdles considering how much easier it is to manage multiple servers. > > With the deployment you have described (which I trimmed out of this > reply), I think you'd be fine running a standalone zookeeper process on > three of your Solr servers, so you won't even need a bunch of extra > hardware. > > When combining a load balancer with SolrCloud, the handler definitions > in solrconfig.xml should set preferLocalShards to true (which Tom > mentioned) so the load balancer target is the machine that actually > processes the request. Troubleshooting becomes more difficult if you > don't do this, and avoiding the extra network hop will help performance. > > Thanks, > Shawn > >