pg.ddx.io  pgsql-committers@postgresql.org mailing list archive  
help / color / mirror / Atom feed
From: Justin Pryzby <pryzby@telsasoft.com>
To: Thomas Munro <thomas.munro@gmail.com>
Cc: Thomas Munro <tmunro@postgresql.org>
Cc: pgsql-committers@lists.postgresql.org
Subject: Re: pgsql: Fix unlink() for STATUS_DELETE_PENDING on Windows.
Date: Tue, 25 Oct 2022 18:57:54 -0500
Message-ID: <20221025235753.GT16921@telsasoft.com> (raw)
In-Reply-To: <CA+hUKGLx+hd9AUy5Cb7TO9osgV84_NT0-ggHy7B6MBpcVpmmLQ@mail.gmail.com>
References: <E1onAdK-000gt6-BV@gemulon.postgresql.org>
	<20221025212102.GS16921@telsasoft.com>
	<20221025213055.GA8537@telsasoft.com>
	<CA+hUKGLx+hd9AUy5Cb7TO9osgV84_NT0-ggHy7B6MBpcVpmmLQ@mail.gmail.com>

On Wed, Oct 26, 2022 at 11:15:16AM +1300, Thomas Munro wrote:
> On Wed, Oct 26, 2022 at 10:31 AM Justin Pryzby <pryzby@telsasoft.com> wrote:
> > On Tue, Oct 25, 2022 at 04:21:02PM -0500, Justin Pryzby wrote:
> > > > Restore that behavior with an explicit check, to see if it fixes the
> > > > occasional 'directory not empty' failures seen in the pg_upgrade tests
> > > > on CI.  Further improvements are possible with proposed upgrades to
> > > > modern Windows APIs that would replace this convoluted code.
> 
> > > If I'm not wrong, this didn't fix the issue you said it fixed.
> >
> > s/said/hoped/sorry
> 
> Drat.  More theories needed then.  Perhaps it has nothing to do with
> my recent changes.  I am starting to wonder if we should have an
> rmdir() wrapper that waits for zombie files to go away, and spits out
> the name of the file that's in the way, and also the name/pid of the
> process that has it open[1] if it times out, so we have a fighting
> chance of debugging this type of stuff.
> 
> [1] https://devblogs.microsoft.com/oldnewthing/20120217-00/?p=8283

FWIW, your description of the problem sounded a bit off to me.

You seemed to say that the problem is with rmdir() on a directory, which
contains a file which was "recently removed", but still opened by
something.

But as I recall, at the point that pg_upgrade is running rmdir(), the
contained files have been unlinked, and the postgres process has ended:

| # Running: pg_upgrade --no-sync -d C:\cirrus\build/testrun/pg_upgrade/002_pg_upgrade\data/t_002_pg_upgrade_old_node_data/pgdata -D C:\cirrus\build/testrun/pg_upgrade/002_pg_upgrade\data/t_002_pg_upgrade_new_node_data/pgdata -b C:/cirrus/build/tmp_install/bin -B C:/cirrus/build/tmp_install/bin -s C:/Users/ContainerAdministrator/AppData/Local/Temp/f67bWmckRH -p 57611 -P 57612 --check
| Performing Consistency Checks
| -----------------------------
| Checking cluster versions                                   ok
| Checking database user is the install user                  ok
| Checking database connection settings                       ok
| Checking for prepared transactions                          ok
| Checking for system-defined composite types in user tables  ok
| Checking for reg* data types in user tables                 ok
| Checking for contrib/isn with bigint-passing mismatch       ok
| Checking for presence of required libraries                 ok
| Checking database user is the install user                  ok
| Checking for prepared transactions                          ok
| Checking for new cluster tablespace directories             ok
| 
| *Clusters are compatible*
| pg_upgrade: warning: could not remove file or directory "C:/cirrus/build/testrun/pg_upgrade/002_pg_upgrade/data/t_002_pg_upgrade_new_node_data/pgdata/pg_upgrade_output.d/20220925T193346.158/log": Directory not empty
| pg_upgrade: warning: could not remove file or directory "C:/cirrus/build/testrun/pg_upgrade/002_pg_upgrade/data/t_002_pg_upgrade_new_node_data/pgdata/pg_upgrade_output.d/20220925T193346.158": Directory not empty

My thinking has always been that this is essentially a bug/deficiency in
the windows FS.  It seems absurd that unlink/rmdir would be
asynchronous.

It'd be swell if we could use a separate device in CI, to be used for
running tests.  Bonus points if it supports COW.

-- 
Justin





view thread (7+ messages)  latest in thread

Message-ID: <20221025235753.GT16921@telsasoft.com>
Permalink:  ../20221025235753.GT16921@telsasoft.com/
Also on:    postgresql.org/message-id/20221025235753.GT16921@telsasoft.com

 · 

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients from your mail client; the link fills in
  the headers below: reply via email
  Some clients, webmail especially, drop In-Reply-To and References from
  a link, and the reply then starts a new thread: use the mbox below.

  To: pgsql-committers@postgresql.org
  Cc: pryzby@telsasoft.com, thomas.munro@gmail.com, tmunro@postgresql.org
  Subject: Re: pgsql: Fix unlink() for STATUS_DELETE_PENDING on Windows.
  In-Reply-To: <20221025235753.GT16921@telsasoft.com>
  References: <E1onAdK-000gt6-BV@gemulon.postgresql.org> <20221025212102.GS16921@telsasoft.com> <20221025213055.GA8537@telsasoft.com> <CA+hUKGLx+hd9AUy5Cb7TO9osgV84_NT0-ggHy7B6MBpcVpmmLQ@mail.gmail.com> <20221025235753.GT16921@telsasoft.com>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox