Usage of MySQL foreign key referencing multiple columns

Question

I just stumbled across possibility of MySQL foreign key to reference multiple columns. I would like to know what is main purpose of multi-column foreign keys like shown bellow

ALTER TABLE `device` 
ADD CONSTRAINT `fk_device_user`
  FOREIGN KEY (`user_created_id` , `user_updated_id` , `user_deleted_id`)
  REFERENCES `user` (`id` , `id` , `id`)
  ON DELETE NO ACTION
  ON UPDATE NO ACTION;

My questions are

1. Is it the same as creating three independent foreign keys?
2. Are there any pros / cons of using one or another?
3. What is the exact use-case for this? (main question)

Answer 1

1. Is it the same as creating three independent foreign keys?

No. Consider the following.

First off, it is not useful to think of it as (id,id,id), but rather (id1,id2,id3) in reality. Because a tuple of (id,id,id) would have no value over just a single column index on id. As such you will see the schema below that depicts that.

create schema FKtest001;
use FKtest001;

create table user
(   id int auto_increment primary key,
    fullname varchar(100) not null,
    id1 int not null,
    id2 int not null,
    id3 int not null,
    index `idkUserTuple` (id1,id2,id3)
);

create table device
(   id int auto_increment primary key,
    something varchar(100) not null,
    user_created_id int not null,
    user_updated_id int not null,
    user_deleted_id int not null,
    foreign key `fk_device_user` (`user_created_id` , `user_updated_id` , `user_deleted_id`)
       REFERENCES `user` (`id1` , `id2` , `id3`)

);
show create table device;
CREATE TABLE `device` (
   `id` int(11) NOT NULL AUTO_INCREMENT,
   `something` varchar(100) NOT NULL,
   `user_created_id` int(11) NOT NULL,
   `user_updated_id` int(11) NOT NULL,
   `user_deleted_id` int(11) NOT NULL,
   PRIMARY KEY (`id`),
   KEY `fk_device_user` (`user_created_id`,`user_updated_id`,`user_deleted_id`),
   CONSTRAINT `device_ibfk_1` FOREIGN KEY (`user_created_id`, `user_updated_id`, `user_deleted_id`) REFERENCES `user` (`id1`, `id2`, `id3`)
 ) ENGINE=InnoDB DEFAULT CHARSET=latin1;
show indexes from device; -- shows 2 indexes (a PK, and composite BTREE)
-- FOCUS heavily on the `Seq_in_index` column for the above

-- xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

drop table device;
drop table user;

create table user
(   id int auto_increment primary key,
    fullname varchar(100) not null,
    id1 int not null,
    id2 int not null,
    id3 int not null,
    index `idkUser1` (id1),
    index `idkUser2` (id2),
    index `idkUser3` (id3)
);

create table device
(   id int auto_increment primary key,
    something varchar(100) not null,
    user_created_id int not null,
    user_updated_id int not null,
    user_deleted_id int not null,
    foreign key `fk_device_user1` (`user_created_id`)
       REFERENCES `user` (`id1`),
    foreign key `fk_device_user2` (`user_updated_id`)
       REFERENCES `user` (`id2`),
    foreign key `fk_device_user3` (`user_deleted_id`)
       REFERENCES `user` (`id3`)
);
show create table device;
CREATE TABLE `device` (
   `id` int(11) NOT NULL AUTO_INCREMENT,
   `something` varchar(100) NOT NULL,
   `user_created_id` int(11) NOT NULL,
   `user_updated_id` int(11) NOT NULL,
   `user_deleted_id` int(11) NOT NULL,
   PRIMARY KEY (`id`),
   KEY `fk_device_user1` (`user_created_id`),
   KEY `fk_device_user2` (`user_updated_id`),
   KEY `fk_device_user3` (`user_deleted_id`),
   CONSTRAINT `device_ibfk_1` FOREIGN KEY (`user_created_id`) REFERENCES `user` (`id1`),
   CONSTRAINT `device_ibfk_2` FOREIGN KEY (`user_updated_id`) REFERENCES `user` (`id2`),
   CONSTRAINT `device_ibfk_3` FOREIGN KEY (`user_deleted_id`) REFERENCES `user` (`id3`)
 ) ENGINE=InnoDB DEFAULT CHARSET=latin1;
 show indexes from device; -- shows 4 indexes (a PK, and 3 indiv FK indexes)
-- FOCUS heavily on the `Seq_in_index` column for the above

There are 2 sections there. The show indexes from device will show the difference of, in the top part, 2 indexes maintained. In the bottom part, 4 indexes maintained. If for some reason the index tuple in the top part is useful for the system, then that tuple approach is certainly the way to go.

The reason is the following. The tuple exists as a group. Think of it as an instance of a set that has meaning as a group. Compare that to the mere existence of the individual parts, and there is a difference. It is not that the users exist, it is that there is a user row that has that tuple as an existence.

2. Are there any pros / cons of using one or another?

The pros were described above in the last paragraph: existence as an actual grouping in the user table as a tuple.

They are apple and oranges and used for different purposes.

3. What is the exact use-case for this? (main question)

A use case would be something that requires the existence of the tuple as a group, as opposed to the existence of the individual items. It is used for what is called compositing. Compositing FK's in particular. See this answer of mine Here as one case.

In short, it is when you want to enforce special hard to think of solutions that require Referential Integrity (RI) at a composited level (groupings) of other entities. Many people think it can't be done so they first think TRIGGER enforcement or front-end Enforcement. Fortunately those use cases can be achieved via the FK Composites thus leaving RI at the db level where it should be (and never at the front-end).

Addendum

Request from OP for a better real life example than the link above.

Consider the following schema:

CREATE SCHEMA testRealLifeTuple;
USE testRealLifeTuple;

CREATE TABLE contacts
(   id INT AUTO_INCREMENT PRIMARY KEY,
    fullname VARCHAR(100) NOT NULL
    -- etc
);

CREATE TABLE tupleHolder
(   -- a tuple representing a necessary Three-some validation
    -- and vetting to get financing
    --
    -- If you can't vett these 3, you can't have my supercomputer financed
    --
    id INT AUTO_INCREMENT PRIMARY KEY,
    CEO INT NOT NULL,   -- Chief Executive Officer
    CFO INT NOT NULL,   -- Chief Financial Officer
    CIO INT NOT NULL,   -- Chief Geek
    creditWorthiness INT NOT NULL, -- 1 to 100. 100 is best

    -- the unique index is necessary for the device FK to succeed
    UNIQUE INDEX `idk_ContactTuple` (CEO,CFO,CIO), -- No duplicates ever. Good for re-use

    FOREIGN KEY `fk_th_ceo` (`CEO`) REFERENCES `contacts` (`id`),
    FOREIGN KEY `fk_th_cfo` (`CFO`) REFERENCES `contacts` (`id`),
    FOREIGN KEY `fk_th_cio` (`CIO`) REFERENCES `contacts` (`id`)
);

CREATE TABLE device
(   -- An Expensive Device, typically our Supercomputer that requires Financing.
    -- This device is so wildly expense we want to limit data changes
    --
    -- Note that the GRANTS (privileges) on this table are restricted.
    --
    id INT AUTO_INCREMENT PRIMARY KEY,
    something VARCHAR(100) NOT NULL,
    CEO INT NOT NULL,   -- Chief Executive Officer
    CFO INT NOT NULL,   -- Chief Financial Officer
    CIO INT NOT NULL,   -- Chief Geek
    FOREIGN KEY `fk_device_2_tuple` (`CEO` , `CFO` , `CIO`)
        REFERENCES `tupleHolder` (`CEO` , `CFO` , `CIO`)
    --
    -- Note that the GRANTS (privileges) on this table are restricted.
    --
);

DROP SCHEMA testRealLifeTuple;

The highlights of this schema come down to the UNIQUE KEY in tupleHolder table, the FK in device, the GRANT restriction (grants not shown), and the fact that the device is shielded from tomfoolery edits in the tupleHolder because of, as mentioned:

• GRANTS
• That the FK must be honored, so the tupleHolder can't be messed with

If the tupleHolder was messed with (the 3 contacts ids), then the FK would be violated.

Said another way, it is NO WAY the same as the device having an FK based on a single column in device, call it [device.badIdea INT], that would FK back to tupleHolder.id.

Also, as mentioned earlier, this differs from merely having the contacts exist. Rather, it matters that the composition of contacts exists, it is a tuple. And in our case the tuple has been vetted, and has a credit worthiness rating, and the id's in that tuple can't be messed with, after a device is bought, unless sufficient GRANTS allow it. And even then, the FK is in place.

It may take 15 minutes for that to sink in, but there is a Huge difference.

I hope this helps.