Re: Injection points: preloading and runtime arguments

On Tue, May 21, 2024 at 04:29:54PM +0500, Andrey M. Borodin wrote:
> Currently I'm working on the test using this
> $creator->query_until(qr/start/, q(
> \echo start
> select injection_points_wakeup('');
> select test_create_multixact();
> ));
>
> I'm fine if instead of injection_points_wakeup('') I'll have to use
> select injection_points_preload('point name');.

Based on our discussion of last week, please find attached the
promised patch set to allow your SLRU tests to work. I have reversed
the order of the patches, moving the loading part in 0001 and the
addition of the runtime arguments in 0002 as we have a use-case for
the loading, nothing yet for the runtime arguments.

I have also come back to the naming, feeling that "preload" was
overcomplicated. So I have used the word "load" instead across the
board for 0001.

Note that the SQL function injection_points_load() does now an
initialization of the shmem area when a process plugs into the module
for the first time, fixing the issue you have mentioned with your SLRU
test. Hence, you should be able to do a load(), then a wait in the
critical section as there would be no memory allocation done when the
point runs. Another thing you could do is to define a
INJECTION_POINT_LOAD() in the code path you're stressing outside the
critical section where the point is run. This should save from a call
to the SQL function. This choice is up to the one implementing the
test, both can be useful depending on what one is trying to achieve.
--
Michael