Hmmmm ... I'm wondering if the Lucene/Solr version changes are a red herring here ... at first blush all of these symptoms sound like invalid cache hits...
: I'm seeing a really weird problem with Solr 1.3. The best match for a : query will not show up with 10 rows, but will show up if I request more, : sometimes 200, sometimes it takes 1000 rows. if two queries that are functionally differnet (and produce different results) are mistakenly considered equivilent you could see this exact behavior ... queryA gets cached, queryB results in a false cache hit and doesn't inclue the highest scoring document that it might if it had been executed w/o caches. increasing the rows param when re-executing queryB still results in a cache hits because of the queryResultWindowSize. stoping/starting solr "fixes" the probem because the caches are empty and queryB is one of the first things tried when the server restarts (before anyone has a chance to run queryA) : Here is the relevant part of solrconfig. Note that we have added a : JaroWinkler fuzzy search, so the dismax specs have extra decoration. ...can you elaborate on your JaroWinkler customizations? is it possible that the Query objects getting generated have hashCode/equals methods that aren't aware of your customizations? -Hoss
