Re: 15beta1 test failure on mips in isolation/expected/stats

Andres Freund <andres(at)anarazel(dot)de> writes:
> On 2022-05-19 21:42:31 -0400, Tom Lane wrote:
>> Even more interesting, the repeatability varies with the settings
>> of max_connections and max_prepared_transactions.

> That is indeed quite odd.

>> At low values (resp. 20 and 0) I've not been able to make it happen at all,
>> but at 100 and 2 it happens circa three times out of four.

> Is this the only permutation that fails?

No, but those values definitely seem to affect the probability of
failure. I just finished a more extensive series of runs, and got:

successes/tries with max_connections/max_prepared_transactions:
5/5 OK with 20/2
5/5 OK with 100/0
3/5 OK with 100/2
4/5 OK with 200/2
5/6 OK with 100/10
5/6 OK with 50/2
6/10 OK with 100/2

It seems like the probability decreases again if I raise either
number further. So I'm now guessing that this is purely a timing
issue and that somehow 100/2 is near the sweet spot for hitting
the timing window. Why those settings would affect pgstats at
all is unclear, though.

regards, tom lane