From: Tom Lane <tgl@sss.pgh.pa.us>
To: Fujii Masao <masao.fujii@oss.nttdata.com>
cc: Andy Fan <zhihuifan1213@163.com>,
        pgsql-hackers <pgsql-hackers@postgresql.org>
Subject: Re: pgbench error: (setshell) of script 0; execution of meta-command failed
In-reply-to: <b0afe844-84b7-4154-ad2c-6763ff84d876@oss.nttdata.com>
References: <87msfz5qv2.fsf@163.com> <87ikqn5em7.fsf@163.com> <e03fc711-8c48-4fb8-ba94-0862c920c30f@oss.nttdata.com> <b0afe844-84b7-4154-ad2c-6763ff84d876@oss.nttdata.com>
Comments: In-reply-to Fujii Masao <masao.fujii@oss.nttdata.com>
	message dated "Fri, 10 Jan 2025 23:39:13 +0900"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <2422777.1736521065.1@sss.pgh.pa.us>
Date: Fri, 10 Jan 2025 09:57:45 -0500
Message-ID: <2422778.1736521065@sss.pgh.pa.us>
Archived-At: <https://www.postgresql.org/message-id/2422778.1736521065%40sss.pgh.pa.us>
Precedence: bulk

Fujii Masao <masao.fujii@oss.nttdata.com> writes:
> Before this commit, pgbench used pqsignal() from port/pqsignal.c
> to set the signal handler for SIGALRM. This version of pqsignal()
> sets SA_RESTART for frontend code, so fgets() in runShellCommand()
> wouldn't return NULL even if SIGALRM arrived during fgets(),
> preventing the reported error.

> On the other hand, currently, pgbench seems to use pqsignal()
> from legacy-pqsignal.c, which doesn't set SA_RESTART for SIGALRM.
> As a result, SIGALRM can interrupt fgets() in runShellCommand()
> and make it return NULL, leading to the reported error.

Oh, interesting.

> I'm not sure if this change was an intentional result of that commit...

Definitely not.  I think we'd better look into how to undo that
effect.

Since legacy-pqsignal is really not supposed to be used by clients
anymore, maybe we could just adjust it to set SA_RESTART for SIGALRM?
The other alternatives I can think of amount to re-introducing
link order dependencies, which would be horrid.

			regards, tom lane