Re: postgres_fdw : altering foreign table not invalidating prepare statement execution plan.

On 2016/10/20 16:27, Ashutosh Bapat wrote:

I wrote:
>>>> Besides
>>>> that, I modified add_rte_to_flat_rtable so that the plan's dependencies
>>>> are
>>>> tracked, but that would lead to tracking the dependencies of unreferenced
>>>> foreign tables in dead subqueries or the dependencies of foreign tables
>>>> excluded from the plan by eg, constraint exclusion. But I thought that
>>>> would be also OK by the same reason as above.

You wrote:
>>> If those unreferenced relations become references because of the
>>> changes in options, we will require those query dependencies to be
>>> recorded. So, if we are recording query dependencies, we should record
>>> the ones on unreferenced relations as well.

I wrote:
>> I mean plan dependencies here, not query dependencies.

> A query dependency also implies plan dependency since changing query
> implies plan changes. So, if query depends upon something, so does the
> plan.

Right.

Sorry, my explanation was not enough, but what I just wanted to mention
is: for unreferenced relations in the plan (eg, foreign tables excluded
from the plan by constraint exclusion), we don't need to record the
dependencies of those relations on FDW-related objects in the plan
dependency list (ie, glob->invalItems), because the changes to
attributes of the FDW-related objects for those relations wouldn't
affect the plan.

I wrote:
>> Having said that, I like the latest version (v6), so I'd vote for marking
>> this as Ready For Committer.

> Marked that way.

Thanks!

Best regards,
Etsuro Fujita