Re: BUG #15804: Assertion failure when using logging_collector with EXEC_BACKEND

Michael Paquier <michael(at)paquier(dot)xyz> writes:
> On Fri, May 17, 2019 at 01:24:43PM -0700, Andres Freund wrote:
>> I don't think a windows-only patch is the same thing as a feature, one
>> that makes debugging easier to boot, that's applicable to all
>> platforms.

> We are discussing here a set of problems which breaks the logging
> collector to work on Windows, requiring mostly a Windows patch. And I
> am not much willing to mess up with the area of shared memory handling
> on Windows a couple of days before a beta.

Yeah. I started digging around this finally, and I agree that the
Windows side of it is a mess --- in particular, on Windows
internal_forkexec() calls pgwin32_ReserveSharedMemoryRegion, which
means we have a hard-wired assumption there that no postmaster child
process is launched till after shared memory exists. EXEC_BACKEND
on Unix will not reproduce that.

At this point I agree with Michael that reverting 57431a911 is the most
prudent thing to do for beta1. While it's probably not *that* hard to
fix, it will require Windows-specific testing. Moreover, the fact that
there are race conditions involved means that one person's testing might
not catch everything. So I do not think we have enough time to get a
trustworthy fix in place by Monday. And we do have other things to
worry about.

(I'm also getting annoyed that Peter, as the original author of this
commit, isn't doing anything about the issue.)

I wouldn't be totally opposed to un-reverting after beta1, if a
better-tested patch emerges. But we don't have such a patch today
and I don't see how we'll get there this weekend.

regards, tom lane