Re: Windows buildfarm members vs. new async-notify isolation test

On 12/5/19 4:37 AM, Amit Kapila wrote:
> On Wed, Dec 4, 2019 at 9:51 PM Andrew Dunstan
> <andrew(dot)dunstan(at)2ndquadrant(dot)com> wrote:
>> On Wed, Dec 4, 2019 at 12:12 AM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>> Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> writes:
>>>> On Tue, Dec 3, 2019 at 10:10 PM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>>>> Hmm ... just looking at the code again, could it be that there's
>>>>> no well-placed CHECK_FOR_INTERRUPTS? Andrew, could you see if
>>>>> injecting one in what 790026972 added to postgres.c helps?
>>>> I also tried to analyze this failure and it seems this is a good bet,
>>>> but I am also wondering why we have never seen such a timing issue in
>>>> other somewhat similar tests. For ex., one with comment (#
>>>> Cross-backend notification delivery.). Do they have a better way of
>>>> ensuring that the notification will be received or is it purely
>>>> coincidental that they haven't seen such a symptom?
>>> TBH, my bet is that this *won't* fix it, but it seemed like an easy
>>> thing to test. For this to fix it, you'd have to suppose that we
>>> never do a CHECK_FOR_INTERRUPTS during a COMMIT command, which is
>>> improbable at best.
>>>
>>
>> You win your bet. Tried this on frogmouth and it still failed.
>>
> IIUC, this means that commit (step l2commit) is finishing before the
> notify signal is reached that session. If so, can we at least confirm
> that by adding something like select pg_sleep(1) in that step? So,
> l2commit will be: step "l2commit" { SELECT pg_sleep(1); COMMIT; }. I
> think we can try by increasing sleep time as well to confirm the
> behavior if required.
>

Yeah, with the sleep in there the NOTIFY is seen.

cheers

andrew

--
Andrew Dunstan https://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services