A foreign key relationship allows you to declare that an index in one table is related to an index in another and allows you to place constraints on what may be done to the table containing the foreign key. The database enforces the rules of this relationship to maintain referential integrity. For example, the score table in the sampdb sample database contains a student_id column, which we use to relate score records to students in the student table. When we created these tables in Chapter 1, we did not set up any explicit relationship between them. Were we to do so, we would declare score.student_id to be a foreign key for the student.student_id column. That prevents a record from being entered into the score table unless it contains a student_id value that exists in the student table. (In other words, the foreign key prevents entry of scores for non-existent students.) We could also set up a constraint such that if a student is deleted from the student table, all corresponding records for the student in the score table should be deleted automatically as well. This is called cascaded delete because the effect of the delete cascades from one table to another. Foreign keys help maintain the consistency of your data, and they provide a certain measure of convenience. Without foreign keys, you are responsible for keeping track of inter-table dependencies and maintaining their consistency from within your applications. In many cases, doing this isn't really that much work. It amounts to little more than adding a few extra DELETE statements to make sure that when you delete a record from one table, you also delete the corresponding records in any related tables. But if your tables have particularly complex relationships, you may not want to be responsible for implementing these dependencies in your applications. Besides, if the database engine will perform consistency checks for you, why not let it? Foreign key support in MySQL is provided by the InnoDB table handler. This section describes how to set up InnoDB tables to define foreign keys, and how foreign keys affect the way you use tables. But first, it's necessary to define some terms:
InnoDB enforces these rules to guarantee that the foreign key relationship stays intact with no mismatches. This is called referential integrity. The syntax for declaring a foreign key in a child table is as follows, with optional parts shown in square brackets: FOREIGN KEY [index_name] (index_columns) REFERENCES tbl_name (index_columns) [ON DELETE action] [ON UPDATE action] [MATCH FULL | MATCH PARTIAL] Note that although all parts of this syntax are parsed, InnoDB does not implement the semantics for all the clauses. The ON UPDATE and MATCH clauses are not supported and are ignored if you specify them.[1]
The parts of the definition that InnoDB pays attention to are:
To define a foreign key, adhere to the following guidelines:
Let's see an example of how all this works. Begin by creating tables named parent and child, such that the child table contains a foreign key that references the par_id column in the parent table: CREATE TABLE parent ( par_id INT NOT NULL, PRIMARY KEY (par_id) ) TYPE = INNODB; CREATE TABLE child ( par_id INT NOT NULL, child_id INT NOT NULL, PRIMARY KEY (par_id, child_id), FOREIGN KEY (par_id) REFERENCES parent (par_id) ON DELETE CASCADE ) TYPE = INNODB; The foreign key in this case uses ON DELETE CASCADE to specify that when a record is deleted from the parent table, child records with a matching par_id value should be removed automatically as well. Now insert a few records into the parent table and add some records that have related key values to the child table: mysql> INSERT INTO parent (par_id) VALUES(1),(2),(3); mysql> INSERT INTO child (par_id,child_id) VALUES(1,1),(1,2); mysql> INSERT INTO child (par_id,child_id) VALUES(2,1),(2,2),(2,3); mysql> INSERT INTO child (par_id,child_id) VALUES(3,1); These statements result in the following table contents, where each par_id value in the child table matches a par_id value in the parent table: mysql> SELECT * FROM parent; +--------+ | par_id | +--------+ | 1 | | 2 | | 3 | +--------+ mysql> SELECT * FROM child; +--------+----------+ | par_id | child_id | +--------+----------+ | 1 | 1 | | 1 | 2 | | 2 | 1 | | 2 | 2 | | 2 | 3 | | 3 | 1 | +--------+----------+ To verify that InnoDB enforces the key relationship for insertion, try adding a record to the child table that has a par_id value not found in the parent table: mysql> INSERT INTO child (par_id,child_id) VALUES(4,1); ERROR 1216: Cannot add a child row: a foreign key constraint fails Now see what happens when you delete a parent record: mysql> DELETE FROM parent where par_id = 1; MySQL deletes the record from the parent table: mysql> SELECT * FROM parent; +--------+ | par_id | +--------+ | 2 | | 3 | +--------+ In addition, it cascades the effect of the DELETE statement to the child table: mysql> SELECT * FROM child; +--------+----------+ | par_id | child_id | +--------+----------+ | 2 | 1 | | 2 | 2 | | 2 | 3 | | 3 | 1 | +--------+----------+ The preceding example shows how to arrange to have deletion of a parent record cause deletion of any corresponding child records. Another possibility is to let the child records remain in the table but have their foreign key columns set to NULL. To do this, it's necessary to make three changes to the definition of the child table:
To see the effect of these changes, recreate the parent table using the original definition and load the same initial records into it. Then create the child table using the new definition shown here: CREATE TABLE child ( par_id INT NULL, child_id INT NOT NULL, UNIQUE (par_id, child_id), FOREIGN KEY (par_id) REFERENCES parent (par_id) ON DELETE SET NULL ) TYPE = INNODB; With respect to inserting new records, the child table behaves the same. That is, it allows insertion of records with par_id values found in the parent table, but disallows entry of values that aren't listed there:[2]
mysql> INSERT INTO child (par_id,child_id) VALUES(1,1),(1,2); mysql> INSERT INTO child (par_id,child_id) VALUES(2,1),(2,2),(2,3); mysql> INSERT INTO child (par_id,child_id) VALUES(3,1); mysql> INSERT INTO child (par_id,child_id) VALUES(4,1); ERROR 1216: Cannot add a child row: a foreign key constraint fails A difference in behavior occurs when you delete a parent record. Try removing a parent record and then check the contents of the child table to see what happens: mysql> DELETE FROM parent where par_id = 1; mysql> SELECT * FROM child; +--------+----------+ | par_id | child_id | +--------+----------+ | NULL | 1 | | NULL | 2 | | 2 | 1 | | 2 | 2 | | 2 | 3 | | 3 | 1 | +--------+----------+ In this case, the child records that had 1 in the par_id column are not deleted. Instead, the par_id column is set to NULL, as specified by the ON DELETE SET NULL constraint. Foreign key capabilities did not all appear at the same time, as shown in the following table. The initial foreign key support prevents insertion or deletion of child records that violate key constraints. The other features were added later.
You can infer from the table that for the most complete foreign key feature support, it's best to use a version of MySQL at least as recent as 3.23.50 if at all possible. Another reason to use more recent versions is that the following problems were not rectified until MySQL 3.23.50:
Living Without Foreign KeysIf you don't have InnoDB support (and thus cannot take advantage of foreign keys), what should you do to maintain the integrity of relationships between your tables? The constraints that foreign keys enforce often are not difficult to implement through application logic. Sometimes, it's simply a matter of how you approach the data entry process. Consider the student and score tables from the grade-keeping project, which are related implicitly through the student_id values in each table. When you administer a test or quiz and have a new set of scores to add to the database, it's unlikely that you'd insert scores for non-existent students. Clearly, the way you'd enter a set of scores would be to start with a list of students from the student table, and then for each one, take the score and use the student's ID number to generate a new score table record. With this procedure, there isn't any possibility of entering a record for a student that doesn't exist, because you wouldn't just invent a score record to put in the score table. What about the case where you delete a student record? Suppose you want to delete student number 13. This also implies you want to delete any score records for that student. With a foreign key relationship in place that specifies cascading delete, you'd simply delete the student table record with the following statement and let MySQL take care of removing the corresponding score table records automatically: DELETE FROM student WHERE student_id = 13; Without foreign key support, you must explicitly delete records for all relevant tables to achieve the same effect as cascading on DELETE: DELETE FROM student WHERE student_id = 13; DELETE FROM score WHERE student_id = 13; Another way to do this, available as of MySQL 4, is to use a multiple-table delete that achieves the same effect as a cascaded delete with a single query. But watch out for a subtle trap. The following statement appears to do the trick, but it's actually not quite correct: DELETE student, score FROM student, score WHERE student.student_id = 13 AND student.student_id = score.student_id; The problem with this statement is that it will fail in the case where the student doesn't have any scores; the WHERE clause will find no matches and thus will not delete anything from the student table. In this case, a LEFT JOIN is more appropriate, because it will identify the student table record even in the absence of any matching score table records: DELETE student, score FROM student LEFT JOIN score USING (student_id) WHERE student.student_id = 13; |