Re: ALTER TABLE lock strength reduction patch is unsafe

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> On Mon, Jun 20, 2011 at 5:56 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> Yeah. This behavior has been there since day zero, and there have been
>> very few complaints about it. But note that there's only a risk for
>> pg_class updates, not any other catalog, and there is exactly one kind
>> of failure with very predictable consequences. The ALTER TABLE patch
>> has greatly expanded the scope of the issue, and that *is* a regression
>> compared to prior releases.

> It's not entirely clear to me how many additional failure cases we've
> bought ourselves with this patch. The particular one you've
> demonstrated seems pretty similar to the on we already had, although
> possibly the window for it is wider.

It's not so much the probability of failure that is bothering me, as
the variety of possible symptoms. There was exactly one failure mode
before, namely "no such relation". I'm not sure how many possible
symptoms there are now, but there's a lot, and most of them are going
to be weird "what the heck was that??" behaviors. If we let 9.1 ship
like this, we are going to be creating a support headache. Even worse,
knowing that those bugs exist will tempt us to write off reports of
weird cache lookup failures as being instances of this problem, when
closer investigation might show that they're something else.

Please note that this position should not be regarded as support for
Simon's proposed patch. I still think the right decision is to revert
the ALTER TABLE feature, mainly because I do not believe this is the
last bug in it. And the fact that there's a pre-existing bug with a
vaguely similar symptom is no justification for introducing more bugs.

regards, tom lane