On Fri, Nov 13, 2015 at 7:15 AM, Merlin Moncure <mmoncure@gmail.com> wrote:
> On Thu, Nov 12, 2015 at 4:26 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Merlin Moncure <mmoncure@gmail.com> writes:
>>> On Thu, Nov 12, 2015 at 9:48 AM, Craig James <cjames@emolecules.com> wrote:
>>>> What about a warning on creation?
>>>>
>>>> db=> create table foo(i integer primary key);
>>>> db=> create table bar(j integer primary key, i integer);
>>>> db=> alter table bar add constraint fk_bar foreign key(i) references foo(i);
>>>> WARNING: fk_bar: column bar(i) has no index, deletions on table foo may be
>>>> slow.
>>>>
>>>> It might save some fraction of these questions.
>>
>>> Maybe, but I wonder if this would cause pg_restore to bleat warnings
>>> when restoring.
>>
>> We could probably teach pg_dump to put index definitions before FKs, if it
>> doesn't already. But I'm suspicious of this sort of "training wheels"
>> warning --- we've had roughly similar messages in the past and removed
>> them because too many people complained about them.
>
> For posterity, indexes are the last step -- and I think that's a good
> way to do things. As to the broader point, I agree. Warnings should
> be reserved for things that are demonstrably dubious, and there are
> just too many situations where that doesn't apply for an unindexed
> foreign constraint. Oh well.
If they were implemented as a notice that would be different. Much
like the notice you get about index creation on PK / Unique constraint
creation. But I'm not 100% sure it's a good idea.