[
https://issues.apache.org/jira/browse/LUCENE-6849?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14991515#comment-14991515
]
Shai Erera commented on LUCENE-6849:
------------------------------------
LGTM. And +1 on making both flush public API. It's an expert API and I believe
users who intend to call {{flush()}} can also understand the implications of
calling {{flush(true, true)}}. Later we can consider consolidating and enhance
this with a {{flush(FlushOptions)}} method, where {{FlushOptions}} lets you
specify whether you want to merge, applyDeletes, segment size flush threshold
etc.
A few comments:
* If you make the second flush() public
** I think we should document perhaps document in ({{flush()}}) when you should
use the second one?
** We should add a testFlushNoCommitButMergeAndApplyDeletes?
** Do you want to also add the second flush() variant to RandomIndexWriter?
* In {{RandomIndexWriter.maybeFlushOrCommit}}, should we also sometimes
randomly apply deletes and trigger merges?
> Add IndexWriter API to write segment(s) without refreshing them
> ---------------------------------------------------------------
>
> Key: LUCENE-6849
> URL: https://issues.apache.org/jira/browse/LUCENE-6849
> Project: Lucene - Core
> Issue Type: New Feature
> Reporter: Michael McCandless
> Assignee: Michael McCandless
> Fix For: Trunk, 5.4
>
> Attachments: LUCENE-6849.patch
>
>
> Today, the only way to have {{IndexWriter}} free up some heap is to invoke
> refresh or flush or close it, but these are all quite costly, and do much
> more than simply "move bytes to disk".
> I think we should add a simple API, e.g. "move the biggest in-memory segment
> to disk" to 1) give more granularity (there could be multiple in-memory
> segments), and 2) only move bytes to disk (not refresh, not fsync, etc.).
> This way apps that want to be more careful on how heap is used can have more
> control.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
