Shamik, The CollapsingQParserPlugin will be available in Solr 4.6 and it should perform much better when collapsing on a high cardinality field. The 4.6 code doesn't directly port back to Solr 4.4 though due to some changes in the build for 4.6. The jira ticket has a conversation about this though and you may be able to follow it and create a patch for 4.4.
Joel On Thu, Oct 31, 2013 at 1:37 AM, Shamik Bandopadhyay <sham...@gmail.com>wrote: > Hi, > > I've recently upgraded to SolrCloud (4.4) from Master-Slave mode. One of > the changes I did the in queries is to add group functionality to remove > duplicate results. The grouping is done on a specific field. But the change > seemed to have a huge effect on the query performance. The "group" option > decreased the performance by 10 times. For e.g. this query takes 1 sec to > execute. The number of results is around 105387. > > > http://localhost:8083/solr/browse?fq=language:(english)&wt=xml&rows=10&start=0&fq=(ContentGroup-local > :"Learn > & Explore" OR ADSKContentGroup-local:"Getting Started")&q=line&sort=score > desc&group=true&group.field=dedup&group.ngroups=true > > If I exclude group option, it comes down to 190ms > > > http://localhost:8083/solr/browse?fq=language:(english)&wt=xml&rows=10&start=0&fq=(ContentGroup-local > :"Learn > & Explore" OR ADSKContentGroup-local:"Getting Started")&q=line > > I'm running this query against a 8 million doc index . I've 2 shard with 1 > replica each, running on a m1x.large EC2 instance, each having 8gb allocat > ed memory. > > Is this a known issue or am I missing something which is making this query > expensive. > > I bumped into this JIRA --> > https://issues.apache.org/jira/browse/SOLR-5027 which > talks about CollapsingQParserPlugin as an alternate to grouping, but that > seemed to be available in 4.6. Just wondering if it can be an alternate in > my case and whether if its possible to apply as a patch in 4.4 version. > > Any pointer will be appreciated. > > - Thanks, > Shamik > -- Joel Bernstein Search Engineer at Heliosearch