On 1 August 2017 at 08:37, Amit Langote <Langote_Amit_f8@lab.ntt.co.jp> wrote:
> On 2017/08/01 15:22, Simon Riggs wrote:
>> On 1 August 2017 at 07:16, Amit Langote <Langote_Amit_f8@lab.ntt.co.jp> wrote:
>>> In f27a6b15e656 (9.6 & later), we decided to "Mark CHECK constraints
>>> declared NOT VALID valid if created with table." In retrospect,
>>> constraints on foreign tables should have been excluded from consideration
>>> in that commit, because the thinking behind the aforementioned commit
>>> (that the constraint is trivially validated because the newly created
>>> table contains no data) does not equally apply to the foreign tables case.
>>>
>>> Should we do something about that?
>>
>> In what way does it not apply? Do you have a failure case?
>
> Sorry for not mentioning the details.
>
> I was thinking that a foreign table starts containing the data of the
> remote object it points to the moment it's created (unlike local tables
> which contain no data to begin with). If a user is not sure whether a
> particular constraint being created in the same command holds for the
> remote data, they may mark it as NOT VALID and hope that the system treats
> the constraint as such until such a time that they mark it valid by
> running ALTER TABLE VALIDATE CONSTRAINT. Since the planner is the only
> consumer of pg_constraint.convalidated, that means the user expects the
> planner to ignore such a constraint. Since f27a6b15e656, users are no
> longer able to expect so.
For Foreign Tables, it sounds like an issue. Don't think it exists for
normal tables.
--
Simon Riggs http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services