Horton

Re: pg_preadv() and pg_pwritev()

Thomas Munro <thomas.munro@gmail.com>

View thread View on postgr.es
From: Thomas Munro <thomas.munro@gmail.com>
To: Andres Freund <andres@anarazel.de>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2021-01-12T23:40:59Z
Lists: pgsql-hackers 
On Mon, Jan 11, 2021 at 3:59 PM Thomas Munro <thomas.munro@gmail.com> wrote:
> On Mon, Jan 11, 2021 at 3:34 PM Thomas Munro <thomas.munro@gmail.com> wrote:
> > I pushed it with that name, and a couple more cosmetic changes.  I'll
> > keep an eye on the build farm.
>
> Since only sifaka has managed to return a result so far (nice CPU), I
> had plenty of time to notice that macOS Big Sur has introduced
> preadv/pwritev.  They were missing on Catalina[1].

The rest of buildfarm was OK with it too, but I learned of a small
problem through CI testing of another patch: it's not OK for
src/port/pwrite.c to do this:

            if (part > 0)
                elog(ERROR, "unexpectedly wrote more than requested");

... because now when I try to use pg_pwrite() in pg_test_fsync,
Windows fails to link:

libpgport.lib(pwrite.obj) : error LNK2019: unresolved external symbol
errstart referenced in function pg_pwritev_with_retry
[C:\projects\postgresql\pg_test_fsync.vcxproj]

I'll go and replace that with an assertion.

Commits

  1. Improve our heuristic for selecting PG_SYSROOT on macOS.

  2. Move our p{read,write}v replacements into their own files.

  3. Don't use elog() in src/port/pwrite.c.

  4. Use vectored I/O to fill new WAL segments.

  5. Provide pg_preadv() and pg_pwritev().