Re: Problem with PostgreSQL 18 and semaphores

[Greg Troxel <gdt%lexort.com@localhost>]

Benny Siegert <bsiegert%netbsd.org@localhost> writes:

> On Sun, 26 Oct 2025, Greg Troxel wrote:
>
>> MESSAGE is abusive here; there is no permanent problem or damage, just
>> pgsql not running.  This belongs in README.NetBSD.  PostgreSQL is very
>> much a huge package where people are expected to read the documentation
>> and we should not promote a culture where people blow that off and then
>> try to save them from themselves.
>
> I would say pgsql not running is very much a permanent problem that
> would warrant having a MESSAGE. Yes, I know the implementation of
> MESSAGE is flawed but still -- we could have both README.NetBSD and
> MESSAGE.

There are lots of other reasons things don't run, especially pgsql.
This is a package where when you change versions you are doing
dump/restore, and it's not like people that don't know what they are
doing can successfully run it anyway.  It is very easy to notice it's
not running, and it's easy to read the logs; that's normal db admin
behavior.

I could also see code in rc.d that on NetBSD sysctls the shm to high
enough when you start the server.  That would actually fix the problem,
and presumably people on a 8 MB VAX where the shm limits would hurt
wouldn't be running pgsql anyay.  (I don't mean to dis vax - I think
it's great that we can run on old arches, and it's good for everyone
that we try.)

I think we're having a problem where this is complicated UNIX and there
are thousands of things that could be wrong, and trying to get shouty
about each one in isolation as Really Important, when that isn't
actually the case.