Re: [PATCH] Fix stale relation close in sequence synchronization

On Fri, 1 May 2026 at 10:45, Ayush Tiwari <ayushtiwari(dot)slg01(at)gmail(dot)com> wrote:
>
> Hi,
>
> On Fri, 1 May 2026 at 09:54, vignesh C <vignesh21(at)gmail(dot)com> wrote:
>>
>> On Fri, 1 May 2026 at 08:58, Ajin Cherian <itsajin(at)gmail(dot)com> wrote:
>> >
>> > This seems to be causing the below buildfarm failure:
>> >
>> > https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=drongo&dt=2026-04-30%2018%3A50%3A23
>>
>> Thanks for reporting this, we are analyzing this. We will analyze and
>> propose a patch for the same.
>>
>
> From what I see it the failure is
> not in the sequence-copy logic itself. It happens before
> the test reaches the privilege check: after the test
> switches the subscription connection to user=regress_seq_repl,
> Windows tries SSPI auth for that login role and rejects it.
>
> I could not run a native Windows SSPI test locally,
> but this follows the existing TAP harness pattern
> used by tests that authenticate as non-default roles.
> The failure occurs because regress_seq_repl is used in
> the subscription connection string but is not passed via
> auth_extra, so pg_regress --config-auth does not add a
> Windows SSPI ident mapping for it.
>
> Meanwhile I'll try to set up a windows machine
> and run the test
>
> meson test -C build subscription/036_sequences --print-errorlogs --verbose

Thanks, the patch worked in my environment. We have made a similar fix
earlier at commit "def0ce3370689b939c6d7a3c3eb824d69989ef6e".
Can you add one comment to say something like:
# Make sure pg_hba.conf is set up to allow connections from regress_seq_repl.
# This is only needed on Windows machines that don't use UNIX sockets.

+$node_publisher->init(
+ allows_streaming => 'logical',
+ auth_extra => [ '--create-role' => 'regress_seq_repl' ]);

Regards
Vignesh