Re: increased duration of stats_ext tests with -DCLOBBER_CACHE_ALWAYS

On 2025-12-06 Sa 7:04 AM, Tomas Vondra wrote:
>
> On 12/6/25 10:00, Alexander Lakhin wrote:
>> Hello Tomas,
>>
>> 03.12.2025 21:23, Tomas Vondra wrote:
>>> On 12/3/25 19:33, Tom Lane wrote:
>>>> I wrote:
>>>>> Yeah, I can imagine that constantly flushing the cached plan for
>>>>> that plpgsql function would be bad. Let me see if I can reformulate
>>>>> that test without using a plpgsql function --- right offhand, it's
>>>>> not obvious why a built-in function wouldn't serve the purpose
>>>>> just as well.
>>>> I pushed a change for this. On my Mac laptop, it brings the time
>>>> for stats_ext with -DCLOBBER_CACHE_ALWAYS down to ~8 minutes, from
>>>> I-didn't-have-the-patience-to-wait-but-it-would-have-been-hours.
>>>>
>>> Thanks!
>> That change decreased the test duration on master:
>> [1] ok 159       + stats_ext                              964154 ms
>> vs
>> [2] ok 157       + stats_ext                            16557572 ms
>>
>> but the test run still failed (and also fails on other branches, where the
>> duration is moderate), probably because of the overall timeout changed
>> somehow.
>>
>> The last good run on avocet, REL_15_STABLE [3] took 12:49:15, however all
>> the failed runs timed out in 4 hours (14400 seconds):
>> timedout [7e54eac] (03:59:27)
>> timedout [7792bdc] (03:59:53)
>> The difference between the last good run and the first bad one I see is:
>> 'script_version' => 'REL_19_1',
>> vs
>> 'script_version' => 'REL_20',
>> though I can't see timeout-related changes in [4], probably something was
>> changed in the environment during the upgrade...
>>
> Yeah, I noticed that too. But I have no idea what could have changed or
> why - the only thing I updated is the buildfarm client :-(
>
> I initially thought it might be due to setting
>
> use_installcheck_parallel => 1
>
> but it still fails after reverting that change. I still have the
> build-farm client v19.1 around, so I can try switching back to that.
>
>

Nothing of significance to what is run changed between 19.1 and 20.

See https://github.com/PGBuildFarm/client-code/compare/REL_19_1...REL_20

cheers

andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com