On 18/11/14 14:49, Manuel Le Normand wrote:
Limiting my answer to the #shards*rows issue, you can have a look at
https://issues.apache.org/jira/browse/SOLR-5611.
Thanks for pointing out SOLR-5611! I did not know about it, and
knowledge about work and ideas in this area was what I wanted to achieve
by this mail to the mailing-list. Havnt dived much into SOLR-5611, but
it seems to me that it will allow you to ask each shard for less than
1000 (if 1000 is the number in the outer super-request) rows in the
get-id-score-sub-requests stage - ask each shard for 1000/#shards (maybe
more?)?
Our solution (as specified in a previous mail) will also "ask each shard
for less than 1000 rows in the get-id-score-sub-request stage". The
difference is that it will start out calculating the exact rows-value to
use for each shard, issuing very in-expensive score-only sub-requests. I
think this approach is at least as nice, if not nicer, as the SOLR-5611
approach.
Regards, Steff
