Re: CommitFest Status Summary - 2010-02-14

Greg Stark <stark@mit.edu>

From: Greg Stark <stark@mit.edu>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Robert Haas <robertmhaas@gmail.com>, pgsql-hackers@postgresql.org
Date: 2010-02-17T12:26:03Z
Lists: pgsql-hackers
On Wed, Feb 17, 2010 at 6:39 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> * Faster CREATE DATABASE by delaying fsync.  This is really two
>> patches now, one of which is apparently to be backpatched.
>
> This one (both parts) seems to have crashed and burned on unexpected
> portability issues :-(.  Do we have any expectation of being able to
> fix it before alpha4?

The current status is that the patch is in HEAD but the one line
fsyncing the directory is #ifdef'd out.
The backpatched fsync of the directory is entirely reverted from early branches.

However I was just looking at the code and realized it has a file
descriptor leak if an error occurs re-opening the files to fsync them.
But then the same file descriptor leak is there if it has an error
opening the destination file so I guess this isn't a new bug.

-- 
greg