jimczi commented on issue #823: LUCENE-8939: Introduce Shared Count Early
Termination In Parallel Search
URL: https://github.com/apache/lucene-solr/pull/823#issuecomment-526120397
> Also, I believe having the abstraction will let future implementations
customize the threshold logic
jimczi commented on issue #823: LUCENE-8939: Introduce Shared Count Early
Termination In Parallel Search
URL: https://github.com/apache/lucene-solr/pull/823#issuecomment-526110754
> So, essentially, we make implicit CollectorManager implementations owned
by IndexSearcher instead of
jimczi commented on issue #823: LUCENE-8939: Introduce Shared Count Early
Termination In Parallel Search
URL: https://github.com/apache/lucene-solr/pull/823#issuecomment-526078472
> That would require a CollectorManager implementation
Not necessarily, you could create a
jimczi commented on issue #823: LUCENE-8939: Introduce Shared Count Early
Termination In Parallel Search
URL: https://github.com/apache/lucene-solr/pull/823#issuecomment-525703089
> Should we commit this then, and then follow up on the other JIRA that I
opened?
Let's benchmark
jimczi commented on issue #823: LUCENE-8939: Introduce Shared Count Early
Termination In Parallel Search
URL: https://github.com/apache/lucene-solr/pull/823#issuecomment-525687970
> Looks like the performance is consistent and we see no degradation. WDYT?
I don't think that these
jimczi commented on issue #823: LUCENE-8939: Introduce Shared Count Early
Termination In Parallel Search
URL: https://github.com/apache/lucene-solr/pull/823#issuecomment-524350989
> Yes, that is correct, but adding it to TopFieldCollector would mean that
it is a part of
jimczi commented on issue #823: LUCENE-8939: Introduce Shared Count Early
Termination In Parallel Search
URL: https://github.com/apache/lucene-solr/pull/823#issuecomment-524333717
Sorry I don't follow. The logic for early termination should be the same in
`TopFieldCollector` than the one