Re: Autovaccuum vs temp tables crash

Magnus Hagander <magnus(at)hagander(dot)net> writes:
> On Fri, Feb 22, 2019 at 7:15 PM Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
>> On Fri, Feb 22, 2019 at 1:14 PM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>> Note that all the temp schemas are made as owned by the bootstrap
>>> superuser, so there is no real argument to be made that people might
>>> be expecting they should be able to delete them.

>> Hmm, well maybe you're right. Just seems like an odd wart.

> Well, the way it works now is you can drop them. But if you then create
> another temp table in the same session, it will get an oid of the already
> dropped schema in the relnamespace column.

Only if you're superuser.

> That just seems plain broken.

There are a *lot* of ways that a superuser can break things. I'm not
real sure that this one is special enough that we need a defense
against it.

However, if someone held a gun to my head and said fix it, I'd be inclined
to do so by having temp-namespace creation insert a "pin" dependency into
pg_depend. Arguably, the only reason we don't create all the temp
namespaces during bootstrap is because we aren't sure how many we'd need
--- but if we did do that, they'd presumably end up pinned.

> I wonder if other "fun" things could happen if you go rename the namespace,
> haven't tried that yet...

I put that one on exactly a par with renaming all the "=" operators.
Yes, the system will let a superuser do it, and no, it's not a good idea.

regards, tom lane