RE: Resetting recovery target parameters in pg_createsubscriber

Hi Hackers,

Thanks for working on the project. I found that one of BF animal fairywren has
been failing [1] the pg_createsubscriber test since the commit 639352d904c.
Can you help solving the issue?

All tests look to be finished but the test runner raised the timeout.
[1] has the output, which may not be included in core. Per my investigation, the
windows OS seems to generate the message.

SUCCESS: The process with PID 8396 (child process of PID 4468) has been terminated.

Also, I found that the node "node_k" cannot be finished at the end of the test.
Because teardown_node()->stop() exits immediately if the $node->{_pid} is not set.
The attribute is set if we start the node by Cluster::start() but node_k is started via
command_ok(pg_ctl...).

So, my theory is that fairywren may wait till all instances are finished at the
end of test. Since one of the nodes missed to be stopped due to an oversight,
it spent 3000 seconds and raised the TIMEOUT. It's bit difficult for me to dig
more because I do not have platform.

Based on above, we can modify how we {start|stop} node_k. Attached patch uses
Utils::system_log() to stop the node.

[1]: https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=fairywren&dt=2026-01-15%2001%3A03%3A12

Best regards,
Hayato Kuroda
FUJITSU LIMITED