On 7/14/2015 2:39 PM, tedsolr wrote: > I am in the process of migrating from a single Solr instance, with multiple > cores, to the SolrCloud. My product uses cores to physically separate our > customers' data: CocaCola has its own core, Pepsi has its own, etc. I want > to keep that physical separation but of course I need horizontal scaling > now. From my readings of the docs, separation in the SolrCloud is provided > by Collections (if a Collection is a logical index). However each Collection > is its own Node. Creating a Node for every customer seems wrong somehow.
There is no requirement for each collection to have its own node. A single collection may require a thousand nodes, or you may have a minimal install where you have only two Solr nodes plus a less-capable third machine for the third zookeeper. In that setup, each collection probably will have one shard and there will be one replica on each of the Solr nodes -- replicationFactor=2. > There are small and large customers - so 5 small customers could share the > same Node. Some customers are big enough to need their own Node(s). So my > question is - can I create multiple Collections within one Node (the > multiple Cores in a single Solr instance model)? If not, are there special > risks to having a cluster with 200 Nodes? The difference seems to be ... is > it better to have one massive JVM per physical server, or lots and lots of > small ones? I personally would only run one Solr instance per node, handling many cores (shard replicas). If you reach a point where you need dozens of gigabytes of heap to handle your indexes, then you might want to think about splitting to multiple solr instances per machine, to keep garbage collection times down ... but in that case, it might make more sense to simply add machines so each JVM heap doesn't need to be as big. > Thanks! > running 4.10 now but considering upgrading to 5.x It's almost always better to run the latest version, especially if you are creating it from the ground up, which is a good idea when trying to migrate to SolrCloud. Thanks, Shawn