It seem the postgres_lock_factory allows multiple of the same lock to be taken out.

For example, if you run this as a unit test, it will fail, because $lock2 will be successfully obtained.

public function test_pg_locks() {

    $factory1 = new \core\lock\postgres_lock_factory('test');

    $lock1 = $factory1->get_lock('myresource', 0);

    $this->assertNotEmpty($lock1);

    $factory2 = new \core\lock\postgres_lock_factory('test');

    $lock2 = $factory2->get_lock('myresource', 0);

    $this->assertFalse($lock2);

    $lock1->release();

 }

To me, if the namespace + resource is the same, the lock should fail to be obtained regardless of the method.

I believe this fails above because of a few preconditions that aren't normally fulfilled:
1. The factory instance being different. This is because the factory does an internal check if it already has the lock (without even going to postgres). Making a new factory instance skips this.

2. The locks being obtained in the same DB session. I'm not super familiar with the internals of pg locks, but I believe if the session is the same postgres will return any advisory locks taken in the same session.

If you change the factory to something else e.g. \core\lock\file_lock_factory, it passes.

My local environment, in case it's relevant:

• Moodle master (currently 2024042200.01)
• PHP 8.1
• Pgbouncer
• Postgres 14