[
https://issues.apache.org/jira/browse/LUCENE-5946?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14133232#comment-14133232
]
Uwe Schindler edited comment on LUCENE-5946 at 9/14/14 2:12 PM:
----------------------------------------------------------------
{quote}
By the way: why did NIO use 16384 bytes bufsize instead of the default 8192? In
my opinion we should consequently use 8 KiB, especially if we do heavy random
access like docvalues.
But this is another discussion for another issue.
{quote}
We no longer use RAF here, so the malloc "bug" in the JVM does not apply. So
16384 is a good choice here, too (was extensively tested for NIOFSDir)! We
should just not make the buffer larger, because NIO internally has a pool of
DirectBuffers to handle the actual disk I/O.
In addition, this is unrelated to the actual buffer size. This is just the
*maximum* to read as chunk if some code requests megabytes of data. The
BufferedIndexInput buffering is unrelated, so most reads (like docvalues) use
much smaller buffers.
was (Author: thetaphi):
{quote}
By the way: why did NIO use 16384 bytes bufsize instead of the default 8192? In
my opinion we should consequently use 8 KiB, especially if we do heavy random
access like docvalues.
But this is another discussion for another issue.
{quote}
We no longer use RAF here, so the malloc "bug" in the JVM does not apply. So
16384 is a good choice here, too (was extensively tested for NIOFSDir)! We
should just not make the buffer larger, because NIO internally has a pool of
DirectBuffers to handle the actual disk I/O.
> Change SimpleFSDirectory to use newByteChannel
> ----------------------------------------------
>
> Key: LUCENE-5946
> URL: https://issues.apache.org/jira/browse/LUCENE-5946
> Project: Lucene - Core
> Issue Type: Bug
> Reporter: Robert Muir
> Attachments: LUCENE-5946.patch
>
>
> Currently our javadocs point to using SimpleFSDirectory "to avoid
> ClosedByInterruptException".
> But this is really bogus. If you interrupt() a thread doing i/o, then you
> need to be prepared for the consequences. Its just that RAF is broken and not
> interruptible at all, but that shouldnt justify us continuing to use it.
> SimpleFSDirectory should be "the most portable", so it should use
> Files.newByteChannel. And we should remove the javadocs/warnings about
> ClosedByInterrupt
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
