Hello Tim,
Since 5.3 you can try to add {!join ... score=none} local parameter, it
switches the algorithm to Lucene's JoinUtil which complexity depends from
number of docs under "from" side.
On Sun, Aug 14, 2016 at 10:13 PM, Tim Frey <tfr...@gmail.com> wrote:
> Hi there. I'm trying to fix a performance problem I have with queries that
> use Solr's Join feature. The query is intended to find all Job
> Applications that have an Interview in a particular state. There are 20
> million Job Applications and around 7 million Interviews, with 1 million
> Interviews in the state I'm looking for. With all other filters applied,
> the total result set is around 5000 documents. The query takes around 10
> seconds.
>
> After reading up on how Joins are essentially just subqueries, I understand
> why my original approach would be slow. However, when I add another
> restriction for the "inner query" to a single Job Application the entire
> query still takes around 5 seconds. In this case, the inner query matches
> 2 documents and the total result set size is 1 document (as expected.)
>
> Here's the debug output:
> https://gist.github.com/tfrey7/50cd92c98e767ec612cc98bf430b9931
>
> I'm using Solr 4.10. All documents are in the same index. The ID columns
> are dynamic integer fields (because we're using the Sunspot ruby library,
> exactly like:
> https://github.com/sunspot/sunspot/blob/master/sunspot_
> solr/solr/solr/configsets/sunspot/conf/schema.xml#L179
> )
>
> Is there something obviously wrong with the query that I'm making? Can
> query-time Joins ever work for a scenario like this?
>
> Thanks!
>
--
Sincerely yours
Mikhail Khludnev
