On 2/8/2016 10:10 PM, Shahzad Masud wrote: > Due to distributed search feature, I might not be able to run > SolrCloud. I would appreciate, if you please share that way of setting > solr home for a specific context in Jetty-Solr. Its good to seek more > information for comparison purposes. Do you think having multiple JVMs > would increase or decrease performance. My document base is around 20 > million rows (in 24 shards), with document size ranging from 100KB - > 400 MB. SM
For most people, the *entire point* of running SolrCloud is to do distributed search, so to hear that you can't run SolrCloud because of distributed search is very confusing to me. I admit to ignorance when it comes to the join feature in Solr ... but it is my understanding that all you need to make joins work properly is to have both of the indexes that you are joining running in the same JVM and the same Solr instance. If you arrange your SolrCloud replicas so a copy of every index is loaded on every server, I think that would satisfy this requirement. I may be wrong, but I believe there are SolrCloud users that use the join feature. When you create a config file for a Solr context, whether it's Jetty, Tomcat, or some other container, you can set the solr/home JNDI variable in the context fragment to set the solr home for that context. I found a specific example for Tomcat. I know Jetty can do the same, but I do not know how to actually create the context fragment. https://wiki.apache.org/solr/SolrTomcat#Installing_Solr_instances_under_Tomcat I need to reiterate one point again. You should only run one Solr container per server, with exactly one Solr context installed in that server. This is recommended whether you're running SolrCloud or not, and whether you're using distributed search or not. One Solr context can handle a LOT of indexes. Running multiple Solr instances per server is only recommended in one case: Extremely large indexes where you would need a very large heap. Running two JVMs with smaller heaps *might* be more efficient ... but in that case, it is usually better to split those indexes between two separate servers, each one running only one instance of Solr. Thanks, Shawn
