Re: postgres_fdw: Oddity in pushing down inherited UPDATE/DELETE joins to remote servers

On 2018/05/11 21:48, Etsuro Fujita wrote:
> (2018/05/11 16:19), Amit Langote wrote:
>> On 2018/05/11 16:12, Amit Langote wrote:
>>> Just to clarify, does this problem only arise because there is a pushed
>>> down join involving the child? That is, does the problem only occur as of
>>> the following commit:
>>>
>>> commit 1bc0100d270e5bcc980a0629b8726a32a497e788
>>> Author: Robert Haas<rhaas(at)postgresql(dot)org>
>>> Date: Wed Feb 7 15:34:30 2018 -0500
>>>
>>> postgres_fdw: Push down UPDATE/DELETE joins to remote servers.
>>>
>>> In other words, do we need to back-patch this up to 9.5 which added
>>> foreign table inheritance?
>>
>> Oops, it should have been clear by the subject line that the problem
>> didn't exist before that commit. Sorry.
>
> No. In theory, I think we could consider this as an older bug added in
> 9.5, because in case of inherited UPDATE/DELETE, the PlannerInfo passed
> to PlanForeignModify doesn't match the one the FDW saw at Path creation
> time, as you mentioned in a previous email, while in case of
> non-inherited UPDATE/DELETE, the PlannerInfo passed to that function
> matches the one the FDW saw at that time. I think that's my fault :(.

Ah, I see. Thanks for clarifying.

> But considering there seems to be no field reports on that, I don't
> think we need back-patching up to 9.5.

Yeah, that might be fine, although it perhaps wouldn't hurt to have the
code match in all branches.

Thanks,
Amit