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

On 2016/04/05 0:23, Tom Lane wrote:
> Amit Langote <amitlangote09(at)gmail(dot)com> writes:
>> On Mon, Apr 4, 2016 at 11:24 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>> A related issue, now that I've seen this example, is that altering
>>> FDW-level or server-level options won't cause a replan either. I'm
>>> not sure there's any very good fix for that. Surely we don't want
>>> to try to identify all tables belonging to the FDW or server and
>>> issue relcache invals on all of them.
>
>> Hm, some kind of PlanInvalItem-based solution could work maybe?
>
> Hm, so we'd expect that whenever an FDW consulted the options while
> making a plan, it'd have to record a plan dependency on them? That
> would be a clean fix maybe, but I'm worried that third-party FDWs
> would fail to get the word about needing to do this.

I would imagine that that level of granularity may be a little too much; I
mean tracking dependencies at the level of individual FDW/foreign
table/foreign server options. I think it should really be possible to do
the entire thing in core instead of requiring this to be made a concern of
FDW authors. How about the attached that teaches
extract_query_dependencies() to add a foreign table and associated foreign
data wrapper and foreign server to invalItems. Also, it adds plan cache
callbacks for respective caches.

One thing that I observed that may not be all that surprising is that we
may need a similar mechanism for postgres_fdw's connection cache, which
doesn't drop connections using older server connection info after I alter
them. I was trying to test my patch by altering dbaname option of a
foreign server but that was silly, ;). Although, I did confirm that the
patch works by altering use_remote_estimates server option. I could not
really test for FDW options though.

Thoughts?

Thanks,
Amit