Horton

Re: Avoid erroring out when unable to remove or parse logical rewrite files to save checkpoint work

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Nathan Bossart <nathandbossart@gmail.com>
Cc: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>, Thomas Munro <thomas.munro@gmail.com>, Tom Lane <tgl@sss.pgh.pa.us>, Andres Freund <andres@anarazel.de>, "Bossart, Nathan" <bossartn@amazon.com>, Julien Rouhaud <rjuju123@gmail.com>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2022-09-02T08:09:14Z
Lists: pgsql-hackers 
On Wed, Aug 31, 2022 at 04:10:59PM +0900, Michael Paquier wrote:
> Yeah, what you have here looks pretty fine to me, so this is IMO
> committable.  Let's wait a bit to see if there are any objections from
> the others, in case.

I had a few hours and I've spent them looking at what you had here in
details, and there were a few things I have tweaked before applying
the patch.  First, elevel was set to LOG for three calls of
get_dirent_type() in code paths where we want to skip entries.  This
would have become very noisy, so I've switched two of them to DEBUG1
and the third one to DEBUG2 for consistency with the surroundings.  A
second thing was RemoveXlogFile(), where we passed down a dirent entry
*and* its d_name.  With this design, it would be easy to mess up
things and pass down a file name that does not match with its dirent
entry, so I have finished by replacing "segname" by the dirent
structure.  A last thing was about the two extra comment blocks in
fd.c, and I could not convince myself that these were helpful hints so
I have removed both of them.
--
Michael

Commits

  1. Expand the use of get_dirent_type(), shaving a few calls to stat()/lstat()

  2. fsync pg_logical/mappings in CheckPointLogicalRewriteHeap().

  3. Introduce logical decoding.