Re: NULL pointer dereference in syslogger with load_libraries() and -DEXEC_BACKEND at startup

On Tue, May 26, 2026, at 8:09 AM, Alvaro Herrera wrote:
> On 2026-May-26, Michael Paquier wrote:
>
>> This issue is different, it is a Postgres logic bug, so adding an
>> exception like the one you are suggesting is just a shortcut hiding
>> the real issue: the log file is not ready yet, but the syslogger is
>> invoked at a point when it thinks the log file exists.
>
> I think we can solve this easily by flipping a new Boolean value at the
> same point were MyBackendType was previously set. The attached POC
> fixes the scenario you described; can you confirm? It needs some
> additional comments, of course.
>

It fixes the issue for me. However, I'm wondering if we can avoid adding a new
boolean for it. We already have redirection_done that guarantees the syslogger
is working properly. Couldn't we use it? (SysLogger_Start() opens the file --
logfile_open -- and a few lines below it sets the redirection_done. If we adopt
this approach, a comment should be added to avoid breaking it in the future.)

> (There is one more place in elog.c where we check that MyBackendType is
> _not_ B_LOGGER, but I think that one is correct as-is; and I'm wondering
> if that would behave correctly before 0c8e082fba8d.)
>

Yes. The redirection_done guarantees that the file is open.

PS> this patch does not allow writes to syslogger as soon as possible like your
patch but it seems acceptable to me.

--
Euler Taveira
EDB https://www.enterprisedb.com/