Re: Injection point locking

On Mon, Jul 08, 2024 at 10:17:49AM -0400, Tom Lane wrote:
> Heikki Linnakangas <hlinnaka(at)iki(dot)fi> writes:
>> Note that until we actually add an injection point to a function that
>> runs in the postmaster, there's no risk. If we're uneasy about that, we
>> could add an assertion to InjectionPointRun() to prevent it from running
>> in the postmaster, so that we don't cross that line inadvertently.

AFAIU, you want to be able to do that to enforce some protocol checks.
That's a fine goal.

> As long as we consider injection points to be a debug/test feature
> only, I think it's a net positive that one can be set in the
> postmaster. I'd be considerably more uncomfortable if somebody
> wanted to do that in production, but maybe it'd be fine even then.

This is documented as a developer feature for tests, the docs are
clear about that.
--
Michael