Something looks wrong... that type of slowdown is certainly not expected. You should be able to see both the main query and a sub-query in the logs... could you post an actual example?
-Yonik http://www.lucidimagination.com On Mon, Jan 4, 2010 at 4:15 AM, Aleksander Stensby <aleksander.sten...@integrasco.com> wrote: > Hi everyone! I've posted a similar question earlier, but in a thread related > to facets in general, so I thought I'd repost it here as a separate thread. > > I have a faceted search that is very fast when I executed the query on a > single solr server, but is significantly slower when executed in a > distributed environment. > The set-back seem to be in the sharding of our data.. And that puzzles me a > little bit... I can't really see why SOLR is so slow at doing this. > > The scenario: > Let's say we have two servers (s1 and s2). > If i query > the following: > q=threadid:33&facet=true&facet.field=author&limit=-1&facet.mincount=0&rows=0 > directly on either server, the response is lightning fast. (<10ms) > > So, in theory I could query them directly, concat the result myself and get > that done pretty fast. > > But if I introduce the shards parameter, the response time booms to between > 15000ms and 20000ms! > shards=s1:8983/solr,s2:8983/solr > > My initial thoughts is that I MUST be doing something wrong here? > > So I try the following: > Run the query on server s1, with the shards param shards=s1:8983/solr > response time goes from sub 10ms to between 5000ms and 10000ms! > Same results if i run the query on s2, and same if i use shards=s2:8983/solr > > Is there really that much overhead in running a distributed facet field > query with Solr? Anyone else experienced this? > > On the other hand, running regular queries without facet distributed is > lightning fast... (so can't really see that this is a network problem or > anything either). - I tried running a facet query on s1 with s1 as the > shards param, and that is still as slow as if the shards param was pointed > to a different server... > > Any insight into this would be greatly appreciated! (Would like to avoid > having to hack together our own solution concatenating results...) > > Cheers, > Aleks >