Just to close the loop. I was fooling around the all the cache setting trying to figure out my problem, so the filterCache is set as part of the experiments. It did not cause any memory issue in this case. After the date rounding adjustment, I re-ran the query with 15 threads with 6000 request and got 1,500/minute throughput by only using a little more than 0.5 GB of Heap Memory. The hit ratio reported in Solr admin statistics page shows filterCache has a hitratio of 0.99. with 103800 lookups and 103773 hits, I assume it is 99%. Have a nice day. -Yao
________________________________ From: Chris Hostetter-3 [via Lucene] [mailto:ml-node+825052-1711725506-201...@n3.nabble.com] Sent: Monday, May 17, 2010 9:04 PM To: Ge, Yao (Y.) Subject: Re: Date faceting and memory leaks : Cache settings: : <filterCache class="solr.LRUCache" size="1512000" initialSize="1512000" : autowarmCount="1280"/> that's a monster filterCache ...i can easly imagine it causing an OOM if your heap is only 5G. : The date rounding suggest is a very good one, I will need to rerun the test : and report back on the cache setting. I remember my filterCache hit ratio is : around 0.7. I did use the tagged results for multi-select display of facet a "hit ratio" or "0.7" ratio, or "0.7% hit rate"? ... with that many unique facet queries, i can't imaging you were getting a 70% hit rate. I'm betting if you monitor that filterCache size and hit rate as you run your test you'll see it just grow and grow until the OOM. and if you analyze the heap dumps you'll probably see the cache hanging on to a ton of DocSets that will never be used again. : values but in this case there is no fq in the load test request URL. I've never tested this, so i can't say for sure, but if it turns out that the filterCache is not your problem, then perhaps there is soemthing wonky with the filterquery exclusion code in cases like this -- where you explicilty exlucde a taged fq but that fq doesn't exist. the qya to rule it out would be to remove the exlcusion from your configs and test it that way to see if the behavior is hte same. -Hoss ________________________________ View message @ http://lucene.472066.n3.nabble.com/Date-faceting-and-memory-leaks-tp8243 72p825052.html To unsubscribe from Re: Date faceting and memory leaks, click here < (link removed) WdlQGZvcmQuY29tfDgyNTAzOHwxNjYwNDQ2MTQ1> . -- View this message in context: http://lucene.472066.n3.nabble.com/Date-faceting-and-memory-leaks-tp824372p825086.html Sent from the Solr - User mailing list archive at Nabble.com.
