I say this only to emphasize that there are always things that are critical that are elided in a testcase that tries to be minimal.
So it seems that there's something critical about my env that I'm failing to tell you all. But what can it be?
Removing permissions also works for me. In my case, I created a brand new PG14.5 cluster in docker and ran Tom's test case.
Your test code seems quite complex to test this simple case...
Guilty as charged. Sorry. I was too focuses on what I wanted to achieve: to stop sessions connected as "r1" killing other sessions connected as "r1".
but looking through it, it appears that you revoked permissions to pg_terminate_backend in the postgres database, but then connected as r1 to the play database, where the permissions weren't revoked. You'll have to revoke the permissions in all databases or change it in template1 before creating the new database.
Yes… that's exactly it. I just replied to that effect here:
