[
https://issues.apache.org/jira/browse/LUCENE-9510?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17195912#comment-17195912
]
Adrien Grand commented on LUCENE-9510:
--------------------------------------
And I just opened a follow-up PR that uses a format that doesn't compress data
for temporary stored fields and term vectors that we write on the fly before
they get sorted upon flushing. On a synthetic benchmark that doesn't index
anything in order to make sure stored fields are the bottleneck, this resulted
in a 3x indexing speedup.
> SortingStoredFieldsConsumer should use a format that has better random-access
> -----------------------------------------------------------------------------
>
> Key: LUCENE-9510
> URL: https://issues.apache.org/jira/browse/LUCENE-9510
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Adrien Grand
> Priority: Minor
> Time Spent: 0.5h
> Remaining Estimate: 0h
>
> We noticed some indexing rate regressions in Elasticsearch after upgrading to
> a new Lucene snapshot. This is due to the fact that
> SortingStoredFieldsConsumer is using the default codec to write stored fields
> on flush. Compression doesn't matter much for this case since these are
> temporary files that get removed on flush after the segment is sorted anyway
> so we could switch to a format that has faster random access.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]