Hi,
On 2026-04-13 09:37:55 +0900, Michael Paquier wrote:
> With this infrastructure already at hand, implementing the
> problematic tests with corrupted varlenas was a matter of minutes,
> leading me to the attached patch (bonus points for check_comprete and
> rawsize)
I think it doesn't scale to have a whole postgres cluster for a test that
takes a few milliseconds. The amount of IO one run of all of postgres' tests
is doing is getting unmangeable, and lots of clusters that are used for a a
second or two that are immediately destroyed contributes substantially to
that.
One PG_TEST_NOCLEAN=1 run with meson ends up with a 33GB testrun/ directory.
And that's without even counting all the pg_regress tests, because there's no
convenient way to disable that. On a smaller machine much of that will be
written to disk due to cache/memory pressure.
There's really no reason for something like this to be a test doing tests via
SQL from what I can tell.
If it does not to be via SSL, can we please start to find a way to combine
tiny stuff like this? We're working hard at making our tests grow
unsustainable.
Greetings,
Andres Freund
