[
https://issues.apache.org/jira/browse/LUCENE-9450?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17179244#comment-17179244
]
Michael McCandless commented on LUCENE-9450:
--------------------------------------------
Whoa, that is good news! It looks like the change is net/net a speedup pure
faceting tasks! Those {{Browse*}} tasks are essentially a
{{MatchAllDocsQuery}} counting facets for all hits.
However, those QPS numbers are silly high and not really trustworthy. Which
{{-source}} did you use? Can you run with {{-source wikimediumall}}? That
will index all ~33.3M documents.
Also, since you had to benchmark two indices (since this change impacts the
index), can you add {{forceMerge = True}} to your two indices? Otherwise each
index might have different segment geometries making the query performance not
very comparable. With {{forceMerge = True}} they will both merge down to one
segment, making the QPS numbers at least comparable if not perhaps realistic in
a production setting.
> Taxonomy index should use DocValues not StoredFields
> ----------------------------------------------------
>
> Key: LUCENE-9450
> URL: https://issues.apache.org/jira/browse/LUCENE-9450
> Project: Lucene - Core
> Issue Type: Improvement
> Components: modules/facet
> Affects Versions: 8.5.2
> Reporter: Gautam Worah
> Priority: Minor
> Labels: performance
> Attachments: wip_taxonomy_patch
>
> Time Spent: 1.5h
> Remaining Estimate: 0h
>
> The taxonomy index that maps binning labels to ordinals was created before
> Lucene added BinaryDocValues.
> I've attached a WIP patch (does not pass tests currently)
> Issue suggested by [~mikemccand]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
