Thread

Re: Random pg_upgrade 004_subscription test failure on drongo

Andrew Dunstan <andrew@dunslane.net> — 2026-05-16T11:22:03Z

On 2026-05-11 Mo 8:10 AM, Andrew Dunstan wrote:
>
> On 2026-05-10 Su 11:00 AM, Alexander Lakhin wrote:
>> Hello Andrew,
>>
>> 13.03.2025 14:19, Andrew Dunstan wrote:
>>>
>>> On 2025-03-13 Th 5:04 AM, vignesh C wrote:
>>>>
>>>> Unfortunately we don't have pg_upgrade_output.d contents in buildfarm
>>>> to see what is the exact reason.
>>>
>>>
>>> That's not supposed to happen. I am testing a fix to see if I can 
>>> make it collect the logs, but for now we'll have to wait till the 
>>> next failure ..
>>
>> Windows animals still produce mysterious pg_upgrade/004_subscription
>> failures: [1] and [2] (and 006_transfer_modes also failed similarly 
>> since
>> then: [3] and [4]). Unluckily, pg_upgrade_output.d/ content is still not
>> shown in the failure logs, so we can only guess why the tests failed. I
>> could not reproduce something alike locally, unfortunately — tried
>> different approaches for several days, but without success, so if we 
>> could
>> see the error/failure reason in the log it would be useful, I hope.
>>
>> Regarding 006_transfer_modes, I thought the reason we have no
>> pg_upgrade_output.d/ is that the test removes it explicitly: [5], but I
>> see no clean_node in 004_subscription, so maybe the log collection is 
>> yet
>> to be fixed in the buildfarm client.
>>
>> Could you have a look, please?
>>
>> [1] 
>> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=fairywren&dt=2026-05-03%2014%3A03%3A12
>> [2] 
>> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=drongo&dt=2026-01-12%2000%3A09%3A37
>> [3] 
>> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=drongo&dt=2025-12-28%2003%3A43%3A24
>> [4] 
>> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=fairywren&dt=2026-01-14%2010%3A52%3A35
>> [5] 
>> https://www.postgresql.org/message-id/61d3ce85-9c7b-4eea-bf83-81272cab00c3%40gmail.com
>>
>>
>
> The test looked ok as it was on drongo and fairywren, but I have 
> loosened it some more on those to see if we can catch the output logs. 
> There are other issues with the 004 test at least (The ok test for the 
> output directory is wrong if there's a failure). But that can wait for 
> now.
>
>
>

OK, I have found and fixed the buildfarm bug. See 
<https://github.com/PGBuildFarm/client-code/commit/55fdf7e0aa355600512a26cdd159b974498f802e>


It will be in the next release, but meanwhile I have deployed it on 
drongo and fairywren.


cheers


andrew


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