Yeah, in this case, I'm running out of memory, and open file descriptors are, I think, just an indicator that IndexSearchers are not getting closed properly. I've already increased the open file descriptors limit, but I'm limited to 2GB of RAM on a 32-bit box.
I'll try explicitly closing searchers now....some number of minutes later - bingo, that takes care of the leak. For the curious, this is what I think happened (the app in question is http://www.simpy.com/ ). The app was burning the CPU like crazy (imagine nearly double-digit load). The IO didn't seem crazy (vmstat/iostat didn't show much bi/bo). Still, after some digging around the source code, I found a place where I was just going through a few thousand Hits on each search against one particular Lucene index. I fixed that problem (nothing to do with IndexSearcher opening/closing), and at the same time I upgraded to Lucene 2.1-dev. This fix doubled the performance of the whole service. Instead of handling N requests/second, it was now handling 2-3*N requests/second. The CPU is still running pretty hot, but the service is handling more search traffic. My guess is that this increase in performance and throughput simply increased the number of different Lucene indices (there are tens of thousands of them per box) being searches, and exposed an issue that was either hidden before, or that the JVM was able to handle before, and now it no longer could. I wish I had a little better grip on what exactly happened here, but I've got a stable application again. Thanks for the help! Otis ----- Original Message ---- From: Yonik Seeley <[EMAIL PROTECTED]> To: java-user@lucene.apache.org Sent: Thursday, November 30, 2006 6:06:50 PM Subject: Re: 2.1-dev memory leak? On 11/30/06, Chris Hostetter <[EMAIL PROTECTED]> wrote: > : IndexSearchers open. The other ones I "let go" without an explicit > : close() call. The assumption is that the old IndexSearchers "expire", > : that they get garbage collected, as I'm no longer holding references to > : them. > > yeah ... that just seems really bad in general, i would try to explicitly > close any searcher your purge from your cache. Yeah, GC may kick off when memory gets low, but the GC system unfortunately doesn't know anything about when file descriptors get low. -Yonik http://incubator.apache.org/solr Solr, the open-source Lucene search server --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
