Hi Torsten, On Tue, Jul 12, 2011 at 2:45 PM, Torsten Kunze <torsten.ku...@b-s-s.de>wrote:
> Hi, > > as a feasibility study I am trying to run Solr with multiple thousands of > cores in the same shard to have small indexes that can be created and > removed very fast. > Now, I have a Tomcat running with 1.600 cores. Memory and open file handles > have been adjusted to be enough for that scenario. > > I am using SolrJ and I implemented a feeder using timer threads to realize > auto commits to each Solr Core independently. > Feeding is done randomly to the cores in parallel. Auto commit is enabled. > > My questions: > Do I need to execute a commit to each core itself or does a commit to one > dedicated core commit all changes of the whole shard? > You need to execute a commit to each core to commit the updates done on it. > Can I feed in parallel to some cores, if a commit or optimize to another > core is currently applied or does Solr block further content integration > requests during that time? > You can feed in parallel and Solr won't block requests to a different core during that time. That being said, your disk would become the bottleneck. > > Because of that many cores, it would be better to define cores for lazy > loading during creation. Unfortunately the current implementation of > CoreAdminHandler does not allow to set the 'loadOnStart' parameter of > solr.xml. Is there a possibility to do this or do I need to implement my own > handlers? > > Does anybody has some good or bad experiences with using many many cores? > I've done a bunch of stuff. There are some (very old) patches as well but probably not useful by themselves. See http://wiki.apache.org/solr/LotsOfCores -- Regards, Shalin Shekhar Mangar.
