From: Tom Lane <tgl@sss.pgh.pa.us>
To: Michael Paquier <michael@paquier.xyz>
cc: pgsql-committers@lists.postgresql.org
Subject: Re: pgsql: Refactor all TAP test suites doing connection checks
In-reply-to: <YGuSe6Q484Eshj2n@paquier.xyz>
References: <E1lTDox-0006rF-OV@gemulon.postgresql.org>
 <513200.1617634642@sss.pgh.pa.us> <YGuSe6Q484Eshj2n@paquier.xyz>
Comments: In-reply-to Michael Paquier <michael@paquier.xyz>
	message dated "Tue, 06 Apr 2021 07:43:07 +0900"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <932058.1617663467.1@sss.pgh.pa.us>
Date: Mon, 05 Apr 2021 18:57:47 -0400
Message-ID: <932059.1617663467@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/932059.1617663467%40sss.pgh.pa.us>
Precedence: bulk

Michael Paquier <michael@paquier.xyz> writes:
> On Mon, Apr 05, 2021 at 10:57:22AM -0400, Tom Lane wrote:
>> I didn't look at the code, but I bet this is similar to the race
>> conditions we've seen before where a server process may exit before
>> the perl test script finishes stuffing a command down the pipe.

> Hmm.  Those failures refer to the 5th and 8th tests of
> 001_password.pl where connect_ok() is used.  The only logical
> differences compared to the previous code are:
> - The addition of on_error_stop in connect_ok() to ignore any errors
> - The fact that this now defines a SQL query (was undef previously).
> - $Test::Builder::Level
> So, the problem is with on_error_stop.

No, I bet it is with "now defines a SQL query".  That means there's
something to be stuffed down the pipe.

I find I can reproduce this pretty reliably on slower machines
here, if you want me to dig into it more carefully.

			regards, tom lane