Re: Use WaitLatch for {pre, post}_auth_delay instead of pg_usleep

On 7/29/21, 12:59 AM, "Kyotaro Horiguchi" <horikyota(dot)ntt(at)gmail(dot)com> wrote:
> At Thu, 29 Jul 2021 09:52:08 +0900, Michael Paquier <michael(at)paquier(dot)xyz> wrote in
>> On Wed, Jul 28, 2021 at 08:28:12PM +0000, Bossart, Nathan wrote:
>> > On 7/28/21, 11:32 AM, "Tom Lane" <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> >> I follow the idea of using WaitLatch to ensure that the delays are
>> >> interruptible by postmaster signals, but even that isn't worth a
>> >> lot given the expected use of these things. I don't see a need to
>> >> expend any extra effort on wait-reporting.
>> >
>> > +1. The proposed patch doesn't make the delay visibility any worse
>> > than what's already there.
>>
>> Agreed to just drop the patch (my opinion about this patch is
>> unchanged). Not to mention that wait events are not available at SQL
>> level at this stage yet.
>
> I'm +1 to not adding wait event stuff at all. So the only advantage
> this patch would offer is interruptivity. I vote +-0.0 for adding that
> interruptivity (+1.0 from the previous opinion of mine:p).

I'm still in favor of moving to WaitLatch() for pre/post_auth_delay,
but I don't think we should worry about the wait reporting stuff. The
patch doesn't add a tremendous amount of complexity, it improves the
behavior on postmaster crashes, and it follows the best practice
described in pgsleep.c of using WaitLatch() for long sleeps.

Nathan