[jira] [Commented] (SOLR-9837) Performance regression of numeric field uninversion time

["Yonik Seeley (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SOLR-9837?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15733368#comment-15733368 ] 

Yonik Seeley commented on SOLR-9837:
------------------------------------

OK, I found the culprit...
https://github.com/apache/lucene-solr/commit/f7aa200d406dbd05a35d6116198302d90b92cb29#diff-595e0e789c5e7ac91fe0300782f1bea6R640

This causes the field to be traversed twice... the first time for docsWithValue, and the second time for the actual uninversion (which also calculates docsWithValue anyway but then doesn't use it).


> Performance regression of numeric field uninversion time
> --------------------------------------------------------
>
>                 Key: SOLR-9837
>                 URL: https://issues.apache.org/jira/browse/SOLR-9837
>             Project: Solr
>          Issue Type: Sub-task
>      Security Level: Public(Default Security Level. Issues are Public) 
>    Affects Versions: master (7.0)
>            Reporter: Yonik Seeley
>            Assignee: Yonik Seeley
>             Fix For: master (7.0)
>
>
> Somehow related to LUCENE-7407, after the transition, the uninvert time of numeric fields has gone up substantially.  I haven't tested all field types yet, just integer fields, which show a 55% performance regression for the initial uninvert time.
> This was tested with a numeric field of cardinality 1M on a 10M doc index.
> {code}
> q=id:1&sort=my_numeric_field desc
> {code}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org