On Mon, Jan 11, 2010 at 7:42 PM, Marc Des Garets <marc.desgar...@192.com>wrote:
> > I am running solr in tomcat and I have about 35 indexes (between 2 and > 80 millions documents each). Currently if I try to update few documents > from an index (let's say the one which contains 80 millions documents) > while tomcat is running and therefore receiving requests, I am getting > few very long garbage collection (about 60sec). I am running tomcat with > -Xms10g -Xmx10g -Xmn2g -XX:PermSize=256m -XX:MaxPermSize=256m. I'm using > ConcMarkSweepGC. > > I have 2 questions: > 1. Is solr doing something specific while an index is being updated like > updating something in memory which would cause the garbage collection? > Solr's caches are thrown away and a fixed number of old queries are re-executed to re-generated the cache on the new index (known as auto-warming). This happens on a commit. > > 2. Any idea how I could solve this problem? Currently I stop tomcat, > update index, start tomcat. I would like to be able to update my index > while tomcat is running. I was thinking about running more tomcat > instance with less memory for each and each running few of my indexes. > Do you think it would be the best way to go? > > If you stop tomcat, how do you update the index? Are you running a multi-core setup? Perhaps it is better to split up the indexes among multiple boxes. Also, you should probably lower the JVM heap so that the full GC pause doesn't make your index unavailable for such a long time. Also see http://www.lucidimagination.com/Community/Hear-from-the-Experts/Articles/Scaling-Lucene-and-Solr -- Regards, Shalin Shekhar Mangar.
