Warming only uses one CPU, so it shouldn't have that much of an impact on a multi-CPU box.
Did this issue begin with Solr 1.3? Perhaps it has something to do with our use of reopen() (to share parts of the index that are not in use). This can lead to greater lock contention while reading from the index. If so, we need 1) an option to disable using IndexReader.reopen() (I think Mark already has a patch for this) 2) NIO support to reduce/eliminate that contention on non Windows platforms (a work in progress - the last patch doesn't actually do it) -Yonik On Wed, Nov 12, 2008 at 2:16 PM, Lance Norskog <[EMAIL PROTECTED]> wrote: > Yes, this is the cache autowarming. > > We turned this off and staged separate queries that pre-warm our standard > queries. We are looking at pulling the query server out of the load balancer > during this process; it is the most effective way to give fixed response > time. > > Lance > > -----Original Message----- > From: oleg_gnatovskiy [mailto:[EMAIL PROTECTED] > Sent: Wednesday, November 12, 2008 11:07 AM > To: solr-user@lucene.apache.org > Subject: Re: Query Performance while updating teh index > > > The rsync seems to have nothing to do with slowness, because while the rsync > is going on, there isn't any reload occurring, once the files are on the > system, it tries a curl request to reload the searcher, which at that point > causes the delays. The file transfer probably has nothing to do with this. > Does this mean that it happens during warming? > > > > Yonik Seeley wrote: >> >> On Tue, Nov 11, 2008 at 9:31 PM, oleg_gnatovskiy >> <[EMAIL PROTECTED]> wrote: >>> Hello. We have an index with 15 million documents working on a >>> distributed environment, with an index distribution setup. While an >>> index on a slave server is being updated, query response times become >>> extremely slow (upwards of 5 seconds). Is there any way to decrease >>> the hit query response times take while an index is being pushed? >> >> Can you tell why it's getting slow? Is this during warming, or does >> it begin during the actual transfer of the new index? >> >> One possibility is that the new index being copied forces out parts of >> the old index from the OS cache. More memory would help in that >> scenario. >> >> -Yonik >> >> > > -- > View this message in context: > http://www.nabble.com/Query-Performance-while-updating-the-index-tp20452835p > 20467099.html > Sent from the Solr - User mailing list archive at Nabble.com. > > >
