[
https://issues.apache.org/jira/browse/LUCENE-6150?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14262526#comment-14262526
]
Uwe Schindler commented on LUCENE-6150:
---------------------------------------
Thanks Mike for confirmation! I will commit this and we will se the impact in
your nightly benchmark, so I think we are fine.
If it is really a problem, I would like to "simplify" the tracking (remove the
onIndexOutputClosed callback - that was the most annoying thing, because it
makes the API horrible, because the tracking is not hidden). In fact, the
tracking on close is not needed here. Its enough to simply add the filename to
the set while opeining the output, then it can be completely private to the
impl in FSDirectory.
> Remove staleFiles set from FSDirectory
> --------------------------------------
>
> Key: LUCENE-6150
> URL: https://issues.apache.org/jira/browse/LUCENE-6150
> Project: Lucene - Core
> Issue Type: Task
> Components: core/store
> Reporter: Uwe Schindler
> Assignee: Uwe Schindler
> Fix For: 5.0, Trunk
>
> Attachments: LUCENE-6150.patch
>
>
> Hi,
> the "hack" to keep track of files written to in FSDirectory, to filter them
> when calling sync is heavily broken. [~mikemccand] already opened an issue,
> which was abandoned then.
> In fact handling this in FSDirectory is a hack and broken! If IndexWriter is
> itsself not able to correctly handle tracking the files, it is also his
> repsonsibilty to do this. We already have a class that can do this:
> TrackingDirectoryWrapper. IndexWriter should use an instance of this class to
> track those stale files (until the problem is solved).
> I would like to keep FSDirectory clean from this, especially, as this is
> broken anyways: If somebody has another directory impl like HDFS or
> Infinispan, the problem still persists. The base directory should throw an
> IOException if trying to sync a file that does not exist!
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
