Re: Partitioning with temp tables is broken

Hi.

On 2018/06/17 22:11, Michael Paquier wrote:
> On Thu, Jun 14, 2018 at 10:38:14PM -0400, Tom Lane wrote:
>> David Rowley <david(dot)rowley(at)2ndquadrant(dot)com> writes:
>>> On 15 June 2018 at 02:42, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>>> I think that if possible, we should still allow a partitioned table
>>>> in which all the rels are temp tables of the current session. What we
>>>> have to disallow is (a) temp/permanent mixes and (b) temp tables from
>>>> different sessions.
>>
>>> So, this used to work in v10. Is it fine to just pluck the feature out
>>> of the v11 release and assume nobody cares?
>>
>> IIUC, it worked in v10 only for small values of "work".
>
> Yeah, if we could get to the set of points mentioned above that would a
> consistent user-facing definition. ATExecAttachPartition() is actually
> heading toward that behavior but its set of checks is incomplete.

Which checks do you think are missing other than those added by the
proposed patch?

> I am quickly looking at forbid-temp-parts-1.patch from previous message
> https://postgr.es/m/a6bab73c-c5a8-2c25-f858-5d6d800a852d@lab.ntt.co.jp
> and this shines per its lack of tests. It would be easy enough to test
> that temp and permanent relations are not mixed within the same session
> for multiple levels of partitioning. Amit, could you add some? There
> may be tweaks needed for foreign tables or such, but I have not looked
> close enough at the problem yet..

OK, I have added some tests. Thanks for looking!

Regards,
Amit