Hi,
On 2014-03-07 13:54:42 -0500, Tom Lane wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
> > I don't think this can actually happen. There are quite a number of
> > things that would go belly-up if you tried to use dynamic shared
> > memory from the postmaster, which is why dsm_create() and dsm_attach()
> > both Assert(IsUnderPostmaster).
>
> Nonetheless it seems like a good idea to make on_exit_reset drop any
> such queued actions.
>
> The big picture here is that in the scenario being debated in the other
> thread, exit() in a child process forked from a backend will execute that
> backend's on_detach actions *even if the code had done on_exit_reset after
> the fork*.
Hm, aren't those actions called via shmem_exit() calling
dsm_backend_shutdown() et al? I think that should be cleared by
on_shmem_exit()?
> So whether or not you buy Andres' argument that it's not
> necessary for atexit_callback to defend against this scenario, there's
> actually no other defense possible given the way things work in HEAD.
I think you're misunderstanding me. I am saying we *should* defend
against it. Our opinions just seem to differ on what to do when the
scenario is detected. I am saying we should scream bloody murder (which
admittedly is a hard in a child), you want to essentially declare it
supported.
Greetings,
Andres Freund
--Andres Freund http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services