When to use "ON UPDATE CASCADE" - sql

Question

I use ON DELETE CASCADE regularly but I never use ON UPDATE CASCADE as I am not so sure in what situation it will be useful.
For the sake of discussion let see some code.
CREATE TABLE parent (
id INT NOT NULL AUTO_INCREMENT,
PRIMARY KEY (id)
);
CREATE TABLE child (
id INT NOT NULL AUTO_INCREMENT, parent_id INT,
INDEX par_ind (parent_id),
FOREIGN KEY (parent_id)
REFERENCES parent(id)
ON DELETE CASCADE
);
For ON DELETE CASCADE, if a parent with an id is deleted, a record in child with parent_id = parent.id will be automatically deleted. This should be no problem.
This means that ON UPDATE CASCADE will do the same thing when id of the parent is updated?
If (1) is true, it means that there is no need to use ON UPDATE CASCADE if parent.id is not updatable (or will never be updated) like when it is AUTO_INCREMENT or always set to be TIMESTAMP. Is that right?
If (2) is not true, in what other kind of situation should we use ON UPDATE CASCADE?
What if I (for some reason) update the child.parent_id to be something not existing, will it then be automatically deleted?
Well, I know, some of the question above can be test programmatically to understand but I want also know if any of this is database vendor dependent or not.
Please shed some light.

Answer 1

It's true that if your primary key is just an identity value auto incremented, you would have no real use for ON UPDATE CASCADE.
However, let's say that your primary key is a 10 digit UPC bar code and because of expansion, you need to change it to a 13-digit UPC bar code. In that case, ON UPDATE CASCADE would allow you to change the primary key value and any tables that have foreign key references to the value will be changed accordingly.
In reference to #4, if you change the child ID to something that doesn't exist in the parent table (and you have referential integrity), you should get a foreign key error.

Answer 2

Yes, it means that for example if you do UPDATE parent SET id = 20 WHERE id = 10 all children parent_id's of 10 will also be updated to 20
If you don't update the field the foreign key refers to, this setting is not needed
Can't think of any other use.
You can't do that as the foreign key constraint would fail.

Answer 3

I think you've pretty much nailed the points!
If you follow database design best practices and your primary key is never updatable (which I think should always be the case anyway), then you never really need the ON UPDATE CASCADE clause.
Zed made a good point, that if you use a natural key (e.g. a regular field from your database table) as your primary key, then there might be certain situations where you need to update your primary keys. Another recent example would be the ISBN (International Standard Book Numbers) which changed from 10 to 13 digits+characters not too long ago.
This is not the case if you choose to use surrogate (e.g. artifically system-generated) keys as your primary key (which would be my preferred choice in all but the most rare occasions).
So in the end: if your primary key never changes, then you never need the ON UPDATE CASCADE clause.
Marc

Answer 4

A few days ago I've had an issue with triggers, and I've figured out that ON UPDATE CASCADE can be useful. Take a look at this example (PostgreSQL):
CREATE TABLE club
(
key SERIAL PRIMARY KEY,
name TEXT UNIQUE
);
CREATE TABLE band
(
key SERIAL PRIMARY KEY,
name TEXT UNIQUE
);
CREATE TABLE concert
(
key SERIAL PRIMARY KEY,
club_name TEXT REFERENCES club(name) ON UPDATE CASCADE,
band_name TEXT REFERENCES band(name) ON UPDATE CASCADE,
concert_date DATE
);
In my issue, I had to define some additional operations (trigger) for updating the concert's table. Those operations had to modify club_name and band_name. I was unable to do it, because of reference. I couldn't modify concert and then deal with club and band tables. I couldn't also do it the other way. ON UPDATE CASCADE was the key to solve the problem.

Answer 5

The ON UPDATE and ON DELETE specify which action will execute when a row in the parent table is updated and deleted. The following are permitted actions : NO ACTION, CASCADE, SET NULL, and SET DEFAULT.
Delete actions of rows in the parent table
If you delete one or more rows in the parent table, you can set one of the following actions:
ON DELETE NO ACTION: SQL Server raises an error and rolls back the delete action on the row in the parent table.
ON DELETE CASCADE: SQL Server deletes the rows in the child table that is corresponding to the row deleted from the parent table.
ON DELETE SET NULL: SQL Server sets the rows in the child table to NULL if the corresponding rows in the parent table are deleted. To execute this action, the foreign key columns must be nullable.
ON DELETE SET DEFAULT: SQL Server sets the rows in the child table to their default values if the corresponding rows in the parent table are deleted. To execute this action, the foreign key columns must have default definitions. Note that a nullable column has a default value of NULL if no default value specified.
By default, SQL Server appliesON DELETE NO ACTION if you don’t explicitly specify any action.
Update action of rows in the parent table
If you update one or more rows in the parent table, you can set one of the following actions:
ON UPDATE NO ACTION: SQL Server raises an error and rolls back the update action on the row in the parent table.
ON UPDATE CASCADE: SQL Server updates the corresponding rows in the child table when the rows in the parent table are updated.
ON UPDATE SET NULL: SQL Server sets the rows in the child table to NULL when the corresponding row in the parent table is updated. Note that the foreign key columns must be nullable for this action to execute.
ON UPDATE SET DEFAULT: SQL Server sets the default values for the rows in the child table that have the corresponding rows in the parent table updated.
FOREIGN KEY (foreign_key_columns)
REFERENCES parent_table(parent_key_columns)
ON UPDATE <action>
ON DELETE <action>;
See the reference tutorial.

Answer 6

It's an excellent question, I had the same question yesterday. I thought about this problem, specifically SEARCHED if existed something like "ON UPDATE CASCADE" and fortunately the designers of SQL had also thought about that. I agree with Ted.strauss, and I also commented Noran's case.
When did I use it? Like Ted pointed out, when you are treating several databases at one time, and the modification in one of them, in one table, has any kind of reproduction in what Ted calls "satellite database", can't be kept with the very original ID, and for any reason you have to create a new one, in case you can't update the data on the old one (for example due to permissions, or in case you are searching for fastness in a case that is so ephemeral that doesn't deserve the absolute and utter respect for the total rules of normalization, simply because will be a very short-lived utility)
So, I agree in two points:
(A.) Yes, in many times a better design can avoid it; BUT
(B.) In cases of migrations, replicating databases, or solving emergencies, it's a GREAT TOOL that fortunately was there when I went to search if it existed.

Answer 7

My comment is mainly in reference to point #3: under what circumstances is ON UPDATE CASCADE applicable if we're assuming that the parent key is not updateable? Here is one case.
I am dealing with a replication scenario in which multiple satellite databases need to be merged with a master. Each satellite is generating data on the same tables, so merging of the tables to the master leads to violations of the uniqueness constraint. I'm trying to use ON UPDATE CASCADE as part of a solution in which I re-increment the keys during each merge. ON UPDATE CASCADE should simplify this process by automating part of the process.

Answer 8

To add to other great answers here it is important to use ON UPDATE CASCADE (or on DELETE CASCADE...) cautiously. Operations on tables with this specification require exclusive lock on underlaying relations.
If you have multiple CASCADE definitions in one table (as in other answer), and especially multiple tables using same definitions, and multiple users updating, this can create a deadlock when one process acquires exclusive lock on first underlaying table, other exclusive lock on second, and they block out each other by none of them being able to get both (all) exclusive locks to perform operation.