[
https://issues.apache.org/jira/browse/LUCENE-8607?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16719130#comment-16719130
]
Alan Woodward commented on LUCENE-8607:
---------------------------------------
Here's the benchmark results using Adrien's idea:
{code:java}
TaskQPS baseline StdDevQPS my_modified_version StdDev Pct diff
MatchAll 503.00 (13.5%) 12164.40 (188.7%) 2318.4% (1864% - 2914%)
{code}
Not quite as much speedup, but it's much simpler in terms of implementation and
is guaranteed not to hit performance when we're collecting all docs while still
being much better than today, so I think this is the way to go.
> Allow MatchAllDocsQuery to skip counting hits
> ---------------------------------------------
>
> Key: LUCENE-8607
> URL: https://issues.apache.org/jira/browse/LUCENE-8607
> Project: Lucene - Core
> Issue Type: Task
> Affects Versions: master (8.0)
> Reporter: Alan Woodward
> Assignee: Alan Woodward
> Priority: Major
> Attachments: LUCENE-8607.patch
>
>
> MatchAllDocsQuery currently uses a private bulk scorer with no
> specialisations for setMinCompetitiveScore(). We've seen what looks to be
> something like a halving of the performance of MatchAllDocsQuery in
> elasticsearch benchmarks running on 8.0 snapshots, and it looks as though
> this is because it's paying the price of keeping track of competitive scores,
> but not actually making use of the new infrastructure. We should modify the
> bulk scorer to early-terminate if setMinCompetitiveScore() is called with a
> value greater than the query's boost.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
