Re: ExecAppendAsyncEventWait() in REL_14_STABLE can corrupt PG_exception_stack

On Sun, Feb 25, 2024 at 6:34 PM Etsuro Fujita <etsuro(dot)fujita(at)gmail(dot)com> wrote:

> > On Fri, Feb 23, 2024 at 01:21:14PM +0300, Alexander Pyhalov wrote:
> > > Recent commit 555276f8594087ba15e0d58e38cd2186b9f39f6d introduced final
> > > cleanup of node->as_eventset in ExecAppendAsyncEventWait().
> > > Unfortunately, now this function can return in the middle of TRY/FINALLY
> > > block, without restoring PG_exception_stack.
> > >
> > > We found this while working on our FDW. Unfortunately, I couldn't reproduce
> > > the issue with postgres_fdw, but it seems it is also affected.
>
> I think this would happen when FDWs configure no events; IIRC I think
> while the core allows them to do so, postgres_fdw does not do so, so
> this would never happen with it.

I was wrong; as you pointed out, this would affect postgres_fdw as
well. See commit 1ec7fca85, which is my commit, but I forgot it
completely. :-(

As I said before, the patch looks good to me. I tweaked comments in
ExecAppendAsyncEventWait() a bit. Attached is an updated patch. In
the patch I also fixed a confusing comment in a related function in
postgres_fdw.c about handling of the in-process request that might be
useless to process.

Sorry, it took more time than expected to get back to this thread.

Best regards,
Etsuro Fujita