BUG #17085: Should be able to create an index without referential checking when ON DELETE NO ACTION

On Tue, Jul 6, 2021 at 3:29 PM PG Bug reporting form <noreply@postgresql.org>
wrote:

The following bug has been logged on the website:

Bug reference: 17085
Logged by: Victor Porton
Email address: porton@narod.ru
PostgreSQL version: 13.3
Operating system: Linux
Description:

When there is either ON DELETE NO ACTION or ON UPDATE NO ACTION for a
foreign key, referential integrity is not preserved anyway.

Yes, it is preserved, the delete or update is prevented.

Therefore in this case ALTER TABLE should be able to create an index

without referential checking.

Indexes are single table only and thus do not care about referential
integrity or perform any referential checking.

It is especially useful for removing superfluous data like:

delete from tags using transactions where not exists(select * from
transactions where tags.tx_id=transactions.id);

I've got into a trouble: This command runs too long because of no foreign
key, but I can't create a foreign key because this command didn't run yet
to
make referntial integrity working.

Foreign keys likewise don't affect performance...they are strictly
concerned with data updates. Now, a FK column is not indexed by default so
depending on how queries are written, and the data in for the FK, there may
be a performance gain from adding an index.

All that is to say if you want to turn a non-FK column into an FK you may
find it helpful to add the index first, fix the data, then add the FK.

David J.

On Wed, Jul 7, 2021 at 12:32 AM David G. Johnston <
david.g.johnston@gmail.com> wrote:

On Tue, Jul 6, 2021 at 3:29 PM PG Bug reporting form <
noreply@postgresql.org> wrote:

The following bug has been logged on the website:

Bug reference: 17085
Logged by: Victor Porton
Email address: porton@narod.ru
PostgreSQL version: 13.3
Operating system: Linux
Description:

When there is either ON DELETE NO ACTION or ON UPDATE NO ACTION for a
foreign key, referential integrity is not preserved anyway.

Yes, it is preserved, the delete or update is prevented.

Therefore in this case ALTER TABLE should be able to create an index

without referential checking.

Indexes are single table only and thus do not care about referential
integrity or perform any referential checking.

It is especially useful for removing superfluous data like:

delete from tags using transactions where not exists(select * from
transactions where tags.tx_id=transactions.id);

I've got into a trouble: This command runs too long because of no foreign
key, but I can't create a foreign key because this command didn't run yet
to
make referntial integrity working.

Foreign keys likewise don't affect performance...they are strictly
concerned with data updates. Now, a FK column is not indexed by default so
depending on how queries are written, and the data in for the FK, there may
be a performance gain from adding an index.

All that is to say if you want to turn a non-FK column into an FK you may
find it helpful to add the index first, fix the data, then add the FK.

David J.

Besides all the good advice by David J., your delete statement has some
issues because it's written more complex than it could.
Transactions is referenced twice, once without correlation and secondly
with correlation.
As a result, if transactions has 0 rows, the statement would delete no rows
at all from tags (although I suppose you'd want to delete them all).
And if transactions is big enough, you may not get an efficient plan.

I suggest you try using this rewrite (besides adding the suggested index):

delete from tags
where not exists (select * from
transactions where tags.tx_id=transactions.id);

Pantelis

On Tue, Jul 6, 2021 at 5:05 PM Pantelis Theodosiou <ypercube@gmail.com>
wrote:

Transactions is referenced twice, once without correlation and secondly

with correlation.

Yeah, the USING clause of DELETE is basically an INNER JOIN. If you need
an ANTI-JOIN you can write that via NOT EXISTS just like you would in a
normal SELECT query. You don't get the option to write an anti-join using
a left join and testing for null.

David J.

On 7/7/21 2:32 AM, David G. Johnston wrote:

On Tue, Jul 6, 2021 at 3:29 PM PG Bug reporting form
<noreply@postgresql.org <mailto:noreply@postgresql.org>> wrote:

The following bug has been logged on the website:

Bug reference:      17085
Logged by:          Victor Porton
Email address: porton@narod.ru <mailto:porton@narod.ru>
PostgreSQL version: 13.3
Operating system:   Linux
Description:

When there is either ON DELETE NO ACTION or ON UPDATE NO ACTION for a
foreign key, referential integrity is not preserved anyway.

Yes, it is preserved, the delete or update is prevented.

Thereforein this case ALTER TABLE should  be able to create an index

without referential checking.

Indexes are single table only and thus do not care about referential
integrity or perform any referential checking.

That's a typo, I meant foreign keys, not indexes.

It is especially useful for removing superfluous data like:

delete from tags using transactions where not exists(select * from
transactions where tags.tx_id=transactions.id
<http://transactions.id&gt;);

I've got into a trouble: This command runs too long because of no
foreign
key, but I can't create a foreign key because this command didn't
run yet to
make referntial integrity working.

Foreign keys likewise don't affect performance...they are strictly
concerned with data updates.  Now, a FK column is not indexed by
default so depending on how queries are written, and the data in for
the FK, there may be a performance gain from adding an index.

All that is to say if you want to turn a non-FK column into an FK you
may find it helpful to add the index first, fix the data, then add the FK.

Anyway, after following my advice PostgreSQL would become more "logical"
and more convenient for users.

There is no reason to have different rules for ALTER TABLE and for
DELETE/UPDATE.

On Tue, Jul 6, 2021 at 5:33 PM Victor Porton <porton.victor@gmail.com>
wrote:

That's a typo, I meant foreign keys, not indexes.

OK, then add the "NOT VALID" qualifier to your ALTER TABLE command. But
you will have to perform validation at some point...

Anyway, after following my advice PostgreSQL would become more "logical"

and more convenient for users.

There is no reason to have different rules for ALTER TABLE and for
DELETE/UPDATE.

It's hard to follow the advice contained in an email that doesn't seem to
understand how things work today.

The subject of the report - which isn't a bug by the way - is to skip
referential checking when adding an index. PostgreSQL doesn't do that
today...

David J.