On Sun, Oct 9, 2011 at 11:30 PM, Jamie Johnson <jej2...@gmail.com> wrote: > I'm doing some work on the solrcloud branch in SVN and am noticing > some strange (but perhaps expected) behavior when executing queries. > I have setup a simple 2 shard cluster, indexed 50 documents into each > (verified by accessing http://localhost:8983/solr/select/?q=*:* and > http://localhost:7574/solr/select/?q=*:* each return 50) but when I > access http://localhost:8983/solr/select/?q=*:*&distrib=true&rows=0 > the total count is 150. If I play with rows, the count actually gets > more accurate as I go, i.e. if I put rows=10, the count is 140, if I > put rows=50 the count is correct (namely 100).
This just sounds like the query is going out to one of the shards twice. Solr is adjusting the total count down based on duplicates received (which is why the count gets more accurate as rows increases). You should probably be able to see this in the logs (sub-requests have isShard=true set). Hopefully this is just a new bug due to all the flux/new code in the branch, and not an existing issue with "distrib=true"? -Yonik http://www.lucene-eurocon.com - The Lucene/Solr User Conference
