[
https://issues.apache.org/jira/browse/LUCENE-9972?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17403914#comment-17403914
]
Michael McCandless commented on LUCENE-9972:
--------------------------------------------
{quote}I guess we could start writing on the cacheDirectory and dynamically
move to an IndexOutput on the FSDirectory if the memory limit is hit?
{quote}
Oh, that's a neat idea. Once we have too many bytes, we would move all already
written bytes off to the wrapped {{Directory}} and delegate all {{writeBytes}}
from that point forwards. Then we don't need to fix all the places in Lucene
that {{createOutput}} to try to "guess" how big the final file might be. I
like it!
> Performance regression in NRTCachingDirectory
> ---------------------------------------------
>
> Key: LUCENE-9972
> URL: https://issues.apache.org/jira/browse/LUCENE-9972
> Project: Lucene - Core
> Issue Type: Bug
> Components: core/store
> Affects Versions: 8.8.2
> Reporter: Markus Gietzen
> Priority: Major
> Labels: performance
>
> This issue has been discussed here:
> [http://mail-archives.apache.org/mod_mbox/lucene-java-user/202105.mbox/%3cam0pr02mb4963712f269c5bf70951f0a3ef...@am0pr02mb4963.eurprd02.prod.outlook.com%3e]
>
> Summary: To get the same speed as in 8.3 I have to overwrite
> NRTCachingDirectory#doCacheWrite
> in 8.8 with the implementation from 8.3. Otherwise it's magnitudes slower.
>
> It seems that the solution for
> https://issues.apache.org/jira/browse/LUCENE-9115 increases the number of
> calls to native-OpenFile (due to falling down to MMAP-Directory) by a very
> large amount.
> I understand the reasoning behind LUCENE-9115 but in fact it destroys the
> performance of our product. For now sub-classing NRTCachingDirectory with the
> old implementation of doCacheWrite() solves the issue but it would be great
> to have an "offical" solution of course.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
