BUG #18741: Detaching a partition referencing a partitioned table fails with a trigger-related error

The following bug has been logged on the website:

Bug reference: 18741
Logged by: Alexander Lakhin
Email address: exclusion(at)gmail(dot)com
PostgreSQL version: 17.2
Operating system: Ubuntu 22.04
Description:

The following script:
CREATE TABLE pt1 (a int PRIMARY KEY) PARTITION BY RANGE (a);
CREATE TABLE p1 PARTITION OF pt1 FOR VALUES FROM (0) TO (1);

CREATE TABLE pt2 (a int) PARTITION BY RANGE (a);
CREATE TABLE p2 (a int, FOREIGN KEY (a) REFERENCES pt1);

ALTER TABLE pt2 ATTACH PARTITION p2 FOR VALUES FROM (0) TO (1);
ALTER TABLE pt2 DETACH PARTITION p2;

fails with:
ERROR: XX000: could not find ON INSERT check triggers of foreign key
constraint 16404
LOCATION: GetForeignKeyCheckTriggers, tablecmds.c:11355

Reproduced on REL_15_STABLE (starting from f4566345c) .. master.

PG Bug reporting form <noreply(at)postgresql(dot)org> 于2024年12月8日周日 22:43写道：

> The following bug has been logged on the website:
>
> Bug reference: 18741
> Logged by: Alexander Lakhin
> Email address: exclusion(at)gmail(dot)com
> PostgreSQL version: 17.2
> Operating system: Ubuntu 22.04
> Description:
>
> The following script:
> CREATE TABLE pt1 (a int PRIMARY KEY) PARTITION BY RANGE (a);
> CREATE TABLE p1 PARTITION OF pt1 FOR VALUES FROM (0) TO (1);
>
> CREATE TABLE pt2 (a int) PARTITION BY RANGE (a);
> CREATE TABLE p2 (a int, FOREIGN KEY (a) REFERENCES pt1);
>
> ALTER TABLE pt2 ATTACH PARTITION p2 FOR VALUES FROM (0) TO (1);
> ALTER TABLE pt2 DETACH PARTITION p2;
>
> fails with:
> ERROR: XX000: could not find ON INSERT check triggers of foreign key
> constraint 16404
> LOCATION: GetForeignKeyCheckTriggers, tablecmds.c:11355
>
> Reproduced on REL_15_STABLE (starting from f4566345c) .. master.
>
>
Thanks for reporting. I can reproduce this on master.

We currently can handle two cases:
1. parent rel has foreign key, but attachrel doesn't have. So we just clone
parent's fks to attachrel.
2. parent rel and attachrel both have foreign keys, and these fks are
consistent, so we can still use attachrel fk not doing expensive clone.

If I understand correctly, we don't do well when parent rel no fks but
attachrel has one.
We can fix above case, but it seems not easy.

Would it work to reject this attaching directly? This may affect users who
already do some attaching, like in this case.
If we allow the above case, we can still attach another partition without a
foreign key. However, maintaining consistent catalog data between parent
and child will take much work.

--
Thanks,
Tender Wang

On Mon, Dec 9, 2024 at 6:15 PM Tender Wang <tndrwang(at)gmail(dot)com> wrote:
>
>
>
> PG Bug reporting form <noreply(at)postgresql(dot)org> 于2024年12月8日周日 22:43写道：
>>
>> The following bug has been logged on the website:
>>
>> Bug reference: 18741
>> Logged by: Alexander Lakhin
>> Email address: exclusion(at)gmail(dot)com
>> PostgreSQL version: 17.2
>> Operating system: Ubuntu 22.04
>> Description:
>>
>> The following script:
>> CREATE TABLE pt1 (a int PRIMARY KEY) PARTITION BY RANGE (a);
>> CREATE TABLE p1 PARTITION OF pt1 FOR VALUES FROM (0) TO (1);
>>
>> CREATE TABLE pt2 (a int) PARTITION BY RANGE (a);
>> CREATE TABLE p2 (a int, FOREIGN KEY (a) REFERENCES pt1);
>>
>> ALTER TABLE pt2 ATTACH PARTITION p2 FOR VALUES FROM (0) TO (1);
>> ALTER TABLE pt2 DETACH PARTITION p2;
>>
>> fails with:
>> ERROR: XX000: could not find ON INSERT check triggers of foreign key
>> constraint 16404
>> LOCATION: GetForeignKeyCheckTriggers, tablecmds.c:11355

>
> Would it work to reject this attaching directly? This may affect users who already do some attaching, like in this case.
> If we allow the above case, we can still attach another partition without a foreign key. However, maintaining consistent catalog data between parent and child will take much work.
>
In the above case, IMO, error out "ALTER TABLE pt2 ATTACH PARTITION p2
FOR VALUES FROM (0) TO (1);"
would be a good option to solve this problem.
otherwise what kind of trigger/constraint will add to pt2?
currently it's none.

To minimize the impact of the change (reducing regression differences),

I think we don't need to error out for the following two cases:
ALTER TABLE <nameX> ATTACH PARTITION <nameY>
1. if the attachee (nameY) already referencing another non-partitioned
table, that should be fine.
2. if the attachee (nameY) referencing another partitioned table and
that partitioned table root table is attacher (nameX)

otherwise, if attachee (nameY) referencing another unrelated
partitioned table, we can error out.