Subject: Re: BUG #16976: server crash when deleting via a trigger on a foreign
 table
To: Thomas Munro <thomas.munro@gmail.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>,
 PostgreSQL mailing lists <pgsql-bugs@lists.postgresql.org>,
 PG Bug reporting form <noreply@postgresql.org>
References: <16976-abe0cb88252bc02e@postgresql.org>
 <756273.1619012414@sss.pgh.pa.us>
 <813ebc08-8f81-4efe-a5f2-caac938b7741@gmail.com>
 <CA+hUKG+XCzNWV82tjgVTUtQ09M6Fuxuow+jCKryPhg+QSo1D_w@mail.gmail.com>
From: Herwig Goemans <herwig.goemans@gmail.com>
Message-ID: <d01e4f40-43e6-310a-6f07-d36096ea564d@gmail.com>
Date: Wed, 28 Apr 2021 14:24:38 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.7.1
MIME-Version: 1.0
In-Reply-To: 
 <CA+hUKG+XCzNWV82tjgVTUtQ09M6Fuxuow+jCKryPhg+QSo1D_w@mail.gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Content-Language: nl-BE
Archived-At: 
 <https://www.postgresql.org/message-id/d01e4f40-43e6-310a-6f07-d36096ea564d%40gmail.com>
Precedence: bulk

Hi Support,


The problem is  in ifx_fdw.c but I do not have dbg packages installed 
for this nor do I know where to find these for ubuntu.

But is there no way you can simulate this issue ? There is a PG server 
crash due to bad user code in a trigger function.

In the mean time we found out that the user code in the trigger function 
was wrong: it has to work with old instead of new values:

      delete from informix.t_water_eenheid

                 where enh_id = old.enh_id;

                 return old;

instead of:


                 delete from informix.t_water_eenheid

                 where enh_id = new.enh_id;

                 return new;

using old it works fine.

But regardless, the PG server should not crash due to bad user code. I 
suppose you agree.

I only work 2 days a week for this client and this is taking up much of 
my time so I was hoping you could simulate this.


Kind regards,


Herwig


On 23/04/2021 00:45, Thomas Munro wrote:
> On Thu, Apr 22, 2021 at 10:09 PM Herwig Goemans
> <herwig.goemans@gmail.com> wrote:
>> And I cannot execute anything in the attached (psql) process,
> A couple of notes about the attach-then-crash approach:  It should be
> "postgres", not "psql", that you attach to, and then you need to type
> "cont" to tell the debugger to allow execution to continue, or it'll
> just block as you described. Sometimes it will stop execution when
> signals arrive, and you can disable that with eg "handle SIGUSR1
> nostop" and then go back to executing with "cont" again.  Once you
> make it crash, you'll be able to use "bt" to see exactly where, but as
> mentioned it'll just be a bunch of big hex numbers unless the symbols
> are installed; at the least it'll show which modules (postgres,
> informix_fdw.so or such).