On Thu, Aug 6, 2020 at 2:22 AM Tom Lane <[email protected]> wrote:
> Robert Haas <[email protected]> writes:
> > On Wed, Aug 5, 2020 at 1:30 PM Tom Lane <[email protected]> wrote:
> >> I'm strongly tempted to convert the trailing Assert to an actual
> >> test-and-elog, too, but didn't do so here.
>
> > I was thinking about that, too. +1 for taking that step.
>
> Will do.
>
> In the longer term, it's annoying that we have no test methodology
> for this other than "manually set a breakpoint here".
One of the methods I see is we can just add some GUC variable for some
action injection. basically it adds some code based on the GUC like this;
if (shall_delay_planning)
{
sleep(10)
};
AFAIK, MongoDB uses much such technology in their test framework. First
it
defines the fail point [1], and then does code injection if the fail point
is set [2].
At last, during the test it can set a fail point like a GUC, but with more
attributes [3].
If that is useful in PG as well and it is not an urgent task, I would like
to help
in this direction.
[1]
https://github.com/mongodb/mongo/search?q=MONGO_FAIL_POINT_DEFINE&unscoped_q=MONGO_FAIL_POINT_DEFINE
[2]
https://github.com/mongodb/mongo/blob/d4e7ea57599b44353b5393afedee8ae5670837b3/src/mongo/db/repl/repl_set_config.cpp#L475
[3]
https://github.com/mongodb/mongo/blob/e07c2d29aded5a30ff08b5ce6a436b6ef6f44014/src/mongo/shell/replsettest.js#L1427
If we're going
> to allow plan-relevant DDL changes to happen with less than full table
> lock, I think we need to improve that. I spent a little bit of time
> just now trying to build an isolationtester case for this, and failed
> completely. So I wonder if we can create some sort of test module that
> allows capture of a plan tree and then execution of that plan tree later
> (even after relcache inval would normally have forced replanning).
> Obviously that could not be a normal SQL-accessible feature, because
> some types of invals would make the plan completely wrong, but for
> testing purposes it'd be mighty helpful to check that a stale plan
> still works.
>
> regards, tom lane
>
>
>
--
Best Regards
Andy Fan
