Re: inherit support for foreign tables

Hi Horiguchi-san,

Thank you for working this patch!

(2014/03/10 17:29), Kyotaro HORIGUCHI wrote:
> Hello. As a minimal implementation, I made an attempt that emit
> NOTICE message when alter table affects foreign tables. It looks
> like following,
>
> | =# alter table passwd add column added int, add column added2 int;
> | NOTICE:  This command affects foreign relation "cf1"
> | NOTICE:  This command affects foreign relation "cf1"
> | ALTER TABLE
> | =# select * from passwd;
> | ERROR:  missing data for column "added"
> | CONTEXT:  COPY cf1, line 1: "root:x:0:0:root:/root:/bin/bash"
> | =#
>
> This seems far better than silently performing the command,
> except for the duplicated message:( New bitmap might required to
> avoid the duplication..

As I said before, I think it would be better to show this kind of
information on each of the affected tables whether or not the affected
one is foreign.  I also think it would be better to show it only when
the user has specified an option to do so, similar to a VERBOSE option
of other commands.  ISTM this should be implemented as a separate patch.

> I made the changes above and below as an attempt in the attached
> patch foreign_inherit-v4.patch
>
>> I think the problem is foreign childs in inheritance tables
>> prevents all menber in the inheritance relation from using
>> parameterized paths, correct?

Yes, I think so, too.

>> Hmm. I tried minimal implementation to do that. This omits cost
>> recalculation but seems to work as expected. This seems enough if
>> cost recalc is trivial here.

I think we should redo the cost/size estimate, because for example,
greater parameterization leads to a smaller rowcount estimate, if I
understand correctly.  In addition, I think this reparameterization
should be done by the FDW itself, becasuse the FDW has more knowledge
about it than the PG core.  So, I think we should introduce a new FDW
routine for that, say ReparameterizeForeignPath(), as proposed in [1].
Attached is an updated version of the patch.  Due to the above reason, I
removed from the patch the message displaying function you added.

Sorry for the delay.

[1] http://www.postgresql.org/message-id/530C7464.6020006@lab.ntt.co.jp

Best regards,
Etsuro Fujita