Horton

Re: Use pg_pwritev_with_retry() instead of write() in dir_open_for_write() to avoid partial writes?

Andres Freund <andres@anarazel.de>

View thread View on postgr.es
From: Andres Freund <andres@anarazel.de>
To: Michael Paquier <michael@paquier.xyz>
Cc: Nathan Bossart <nathandbossart@gmail.com>, Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>, Thomas Munro <thomas.munro@gmail.com>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2022-10-28T22:07:00Z
Lists: pgsql-hackers 
Hi,

On 2022-10-28 11:09:38 +0900, Michael Paquier wrote:
> On Thu, Oct 27, 2022 at 03:58:25PM -0700, Andres Freund wrote:
> > The block sizes don't need to match, do they? As long as the block is properly
> > aligned, we can change the iov_len of the final iov to match whatever the size
> > is being passed in, no?
> 
> Hmm.  Based on what Bharath has written upthread, it does not seem to
> matter if the size of the aligned block changes, either:
> https://www.postgresql.org/message-id/CALj2ACUccjR7KbKqWOsQmqH1ZGEDyJ7hH5Ef+DOhcv7+kOnjCQ@mail.gmail.com
> 
> I am honestly not sure whether it is a good idea to make file_utils.c
> depend on one of the compile-time page sizes in this routine, be it
> the page size of the WAL page size, as pg_write_zeros() would be used
> for some rather low-level operations.  But we could as well just use a
> locally-defined structure with a buffer at 4kB or 8kB and call it a
> day?

Shrug. I don't think we gain much by having yet another PGAlignedXYZBlock.

Greetings,

Andres Freund

Commits

  1. Revise pg_pwrite_zeros()

  2. Use pg_pwrite_zeros() in walmethods.c

  3. Introduce pg_pwrite_zeros() in fileutils.c

  4. Move pg_pwritev_with_retry() to src/common/file_utils.c

  5. Restore pg_pread and friends.

  6. Remove dead pread and pwrite replacement code.