Create and manage foreign key relationships

This page describes how to manage foreign key relationships in your database.

A foreign key is a column that is shared between tables to establish a link between related data. When you use a foreign key, Spanner ensures that this relationship is maintained.

The following diagram shows a basic database schema where data in a table has a relationship to data in another table.

Example database schema showing foreign key relationships between tables.

Figure 1. Diagram of an order processing database schema

There are three tables in the schema shown in Figure 1:

The Customers table records the names of each customer.
The Orders tables keeps track of all orders made.
The Products table stores the product information for every product.

There are two foreign key relationships between these tables:

A foreign key relationship is defined between the Orders table and the Customers table to ensure that an order can't be created unless there is a corresponding customer.
A foreign key relationship between the Orders table and the Products table ensures that an order can't be created for a product that doesn't exist.

Using the previous schema as an example, this topic discusses the Data Definition Language (DDL) CONSTRAINT statements that you can use to manage relationships between tables in a database.

By default, all foreign keys in Spanner are enforced foreign keys, which enforce referential integrity. In Spanner you can also choose to use informational foreign keys, which don't validate or enforce referential integrity. For more information, see Comparison of foreign keys and Choose which foreign key type to use. When unspecified, the foreign keys in examples on this page are enforced foreign keys.

Add a foreign key to a new table

Assume that you've created a Customers table in your basic product ordering database. You now need an Orders table to store information about the orders that customers make. To ensure all orders are valid, you don't want to let the system insert rows into the Orders table unless there's also a matching entry in the Customers table. Therefore, you need an enforced foreign key to establish a relationship between the two tables. One choice is to add a CustomerID column to the new table and use it as the foreign key to create a relationship with the CustomerID column in the Customers table.

When you create a new table with a foreign key, you use REFERENCE to establish a relationship to another table. The table that contains the REFERENCE statement is called the referencing table. The table named in the REFERENCE statement is the referenced table. The column that is named in the REFERENCE statement is called the referencing column.

The following example shows how to use the CREATE TABLE DDL statement to create the Orders table with a foreign key constraint that references CustomerID in the Customers table.

GoogleSQL

CREATE TABLE Orders (
  OrderID INT64 NOT NULL,
  CustomerID INT64 NOT NULL,
  Quantity INT64 NOT NULL,
  ProductID INT64 NOT NULL,
  CONSTRAINT FK_CustomerOrder FOREIGN KEY (CustomerID) REFERENCES Customers (CustomerID)
) PRIMARY KEY (OrderID);

PostgreSQL

CREATE TABLE Orders (
  OrderID BIGINT NOT NULL,
  CustomerID BIGINT NOT NULL,
  Quantity BIGINT NOT NULL,
  ProductID BIGINT NOT NULL,
  CONSTRAINT FK_CustomerOrder FOREIGN KEY (CustomerID) REFERENCES Customers (CustomerID),
  PRIMARY KEY (OrderID)
);

The previous statement contains a CONSTRAINT clause that has the following characteristics:

Use of the CONSTRAINT syntax to name a constraint, making it easier to drop the table using the name you've chosen.
The constraint has the name FK_CustomerOrder. Constraint names are scoped to the schema and must be unique within the schema.
The Orders table, on which you define the constraint, is the referencing table. The Customers table is the referenced table.
The referencing column in the referencing table is CustomerID. It references the CustomerID field in the Customers table. If someone tries to insert a row into Orders with a CustomerID that doesn't exist in Customers, the insert fails.

The following example shows an alternative table creation statement. Here, the foreign key constraint is defined without a name. When you use this syntax, Spanner generates a name for you. To discover the names of all foreign keys, refer to View properties of a foreign key relationship.

GoogleSQL

CREATE TABLE Orders (
  OrderID INT64 NOT NULL,
  CustomerID INT64 NOT NULL,
  ProductID INT64 NOT NULL,
  FOREIGN KEY (CustomerID) REFERENCES Customers (CustomerID)
) PRIMARY KEY (OrderID);

PostgreSQL

CREATE TABLE Orders (
  OrderID BIGINT NOT NULL,
  CustomerID BIGINT NOT NULL,
  Quantity BIGINT NOT NULL,
  ProductID BIGINT NOT NULL,
  FOREIGN KEY (CustomerID) REFERENCES Customers (CustomerID),
  PRIMARY KEY (OrderID)
);

Add a foreign key to an existing table

You also want to make sure that customers can only order products that exist. If your table has existing constraints, you must drop all of the constraints. In Spanner, all enforced constraints in a table must be implemented at the same time in a single batch DDL statement.

If your table has no existing constraints, you can use the ALTER TABLE DDL statement to add an enforced foreign key constraint to the existing Orders table as shown in the following example:

ALTER TABLE Orders
  ADD CONSTRAINT DB_ProductOrder FOREIGN KEY (ProductID) REFERENCES Products (ProductID);

The referencing column in Orders is ProductID, and it references the ProductID column in Products. If you are fine with Spanner naming these constraints for you, use the following syntax:

ALTER TABLE Orders
  ADD FOREIGN KEY (ProductID) REFERENCES Products (ProductID);

Add a foreign key with a delete action to a new table

Recall the previous example where you have a Customers table in a product ordering database that needs an Orders table. You want to add a foreign key constraint that references the Customers table. However, you want to ensure that when you delete a customer record in the future, Spanner also deletes all orders for that customer. In this case, you want to use the ON DELETE CASCADE action with the foreign key constraint.

The following CREATE TABLE DDL statement for the Orders table includes the foreign key constraint that references the Customers table with an ON DELETE CASCADE action.

GoogleSQL

CREATE TABLE Orders (
  OrderID INT64 NOT NULL,
  CustomerID INT64 NOT NULL,
  Quantity INT64 NOT NULL,
  ProductID INT64 NOT NULL,
  CONSTRAINT FK_CustomerOrder FOREIGN KEY (CustomerID)
    REFERENCES Customers (CustomerID) ON DELETE CASCADE
) PRIMARY KEY (OrderID);

PostgreSQL

CREATE TABLE Orders (
  OrderID BIGINT NOT NULL,
  CustomerID BIGINT NOT NULL,
  Quantity BIGINT NOT NULL,
  ProductID BIGINT NOT NULL,
  FOREIGN KEY (CustomerID)
    REFERENCES Customers (CustomerID) ON DELETE CASCADE,
  PRIMARY KEY (OrderID)
);

The previous statement contains a foreign key constraint with an ON DELETE CASCADE clause. The CustomerID column is a foreign key that references the CustomerID field in the Customers table. This means that each CustomerID value in the Orders table must also exist in the Customers table. If someone tries to delete a row from the Customers table, all of the rows in the Orders table that reference the deleted CustomerID value are also deleted in the same transaction.

Add a foreign key with a delete action to a table

You also want to make sure that orders are only created for products that exist. You can use ALTER TABLE to add another foreign key constraint with ON DELETE CASCADE action to the orders table as follows:

ALTER TABLE Orders
  ADD CONSTRAINT DB_ProductOrder FOREIGN KEY (ProductID)
    REFERENCES Products (ProductID) ON DELETE CASCADE;

Deleting a row from the Products table deletes all of the rows in the Orders table that reference the deleted ProductID value.

Use informational foreign keys (GoogleSQL only)

Informational foreign keys let the query optimizer make use of the foreign key relationship without the overhead incurred from referential integrity checks performed by enforced foreign keys. Informational foreign keys are useful when enforcing strict referential integrity is either impractical or incurs significant performance overhead.

Continuing with the previous example, imagine you want to model the relationships between the Customers, Orders, and Products tables. However, enforcing strict referential integrity in the data of the tables might introduce performance bottlenecks, especially during peak shopping periods with high order volumes. Additionally, customers might place orders for products that were discontinued and removed from the Products table.

You can create the Orders table using informational foreign keys:

CREATE TABLE Orders (
    OrderID INT64 NOT NULL,
    CustomerID INT64 NOT NULL,
    Quantity INT64 NOT NULL,
    ProductID INT64 NOT NULL,
    CONSTRAINT FK_CustomerOrder FOREIGN KEY (CustomerID) REFERENCES Customers (CustomerID) NOT ENFORCED,
    CONSTRAINT FK_ProductOrder FOREIGN KEY (ProductID) REFERENCES Products (ProductID) NOT ENFORCED
) PRIMARY KEY (OrderID);

By creating an informational foreign key with NOT ENFORCED, you allow for the possibility that an order might reference a non-existent customer or product. Using an informational foreign key instead of an enforced foreign key constraint is a good choice if a customer account might be deleted or a product might be discontinued. With an informational foreign key, Spanner doesn't perform referential integrity validation. This reduces the write overhead, potentially improving performance during peak order processing times.

You can allow the query optimizer to use the relationships to generate efficient query plans. This can improve the performance of queries that join the tables on foreign key columns. For more information, see informational foreign key for query optimization.

Query data across foreign key relationships

SELECT * FROM Orders
  INNER JOIN Customers ON Orders.CustomerID = Customers.CustomerID
  INNER JOIN Products ON Orders.ProductsID = Products.ProductID;

Referential integrity with enforced foreign keys

The main reason for adding enforced foreign key relationships is so that Spanner can maintain the referential integrity of your data. If you modify data in a way that breaks a foreign key constraint, the update fails with an error.

Consider the data in Figure 2. Some customers have ordered products, as shown in the Orders table. Because of the enforced foreign key constraint that are in place, the data that was inserted into the Orders table has referential integrity.

Sample data for the `Customers`, `Products` and `Orders` tables.

Figure 2. Sample data for in our ordering database.

The following examples show what happens when you try to modify the data in a way that would break referential integrity.

Add a row into the Orders table with a CustomerID value that does not exist in Customers

What happens if you try the following modification, given the sample data from the preceding diagram?
```
INSERT INTO Orders (OrderID, ProductID, Quantity, CustomerID)
  VALUES (19, 337876, 4, 447);
```
In this case, the system would try to insert a row into Orders with a CustomerID (447) that doesn't exist in the Customers table. If the system did this, you would have an invalid order in your system. However, with the enforced foreign key constraint you added to your Orders table, your table is protected. The INSERT fails with the following message, assuming the constraint is called FK_CustomerOrder.
```
Foreign key constraint `FK_CustomerOrder` is violated on table `Orders`.
Cannot find referenced values in Customers(CustomerID).
```
Unlike enforced foreign keys, informational foreign keys don't enforce referential integrity. If FK_CustomerOrder is an informational foreign key, then the insert statement succeeds because Spanner doesn't validate that the corresponding CustomerID exists in the Customers table. Because of this, the data might not conform to the referential integrity defined by FK_CustomerOrder.
Attempt to delete a row from the Customers table when the customer is referenced in an enforced foreign key constraint.

Imagine a situation where a customer unsubscribes from your online store. You want to remove the customer from your backend, so you attempt the following operation.
```
DELETE FROM Customers WHERE CustomerID = 721;
```
In this example, Spanner detects through the foreign key constraint that there are still records in the Orders table that reference the customer row you are trying to delete. The following error is displayed in this case.

Foreign key constraint violation when deleting or updating referenced row(s): referencing row(s) found in table `Orders`.

To fix this issue, you delete all referencing entries in Orders first. You can also define the foreign key with the ON DELETE CASCADE action to let Spanner handle deletion of referencing entries.

Similarly, if FK_CustomerOrder is an informational foreign key, then the delete action succeeds because Spanner doesn't guarantee the referential integrity of informational foreign keys.

View properties of a foreign key relationship

Spanner's INFORMATION_SCHEMA contains information about foreign keys and their backing indexes. The following are some examples of the questions you can answer by querying the INFORMATION SCHEMA.

For more information on backing indexes, see Foreign keys backing indexes.

What constraints are defined in my database?

SELECT tc.CONSTRAINT_NAME, tc.TABLE_NAME, tc.CONSTRAINT_TYPE
FROM INFORMATION_SCHEMA.TABLE_CONSTRAINTS as tc
WHERE tc.CONSTRAINT_TYPE = 'FOREIGN KEY';

What foreign keys are defined in my database?

SELECT rc.CONSTRAINT_NAME, rc.UNIQUE_CONSTRAINT_NAME, rc.SPANNER_STATE
FROM INFORMATION_SCHEMA.REFERENTIAL_CONSTRAINTS as rc;

Which indexes are secondary indexes for foreign keys, also known as backing indexes?

Foreign key backing indexes are managed by Spanner , so querying for SPANNER_IS_MANAGED on the INDEXES view returns all backing indexes.

SELECT i.TABLE_NAME, i.INDEX_NAME, i.INDEX_TYPE, i.INDEX_STATE,
  i.IS_UNIQUE, i.IS_NULL_FILTERED, i.SPANNER_IS_MANAGED
FROM INFORMATION_SCHEMA.INDEXES as i
WHERE SPANNER_IS_MANAGED = 'YES';

What is the referential action defined with the foreign key constraint?

SELECT rc.CONSTRAINT_NAME, rc.UNIQUE_CONSTRAINT_NAME, rc.DELETE_RULE,
  rc.UPDATE_RULE
FROM INFORMATION_SCHEMA.REFERENTIAL_CONSTRAINTS as rc;

Is a foreign key enforced or not enforced?

SELECT tc.CONSTRAINT_NAME, tc.TABLE_NAME, tc.CONSTRAINT_TYPE, tc.ENFORCED
FROM INFORMATION_SCHEMA.TABLE_CONSTRAINTS as tc
WHERE tc.CONSTRAINT_TYPE = 'FOREIGN KEY';

For more information, see Information Schema.

Remove a foreign key relationship

The following DDL drops a foreign key constraint from the Orders table.

ALTER TABLE Orders
  DROP CONSTRAINT FK_CustomerOrder;

The foreign key backing indexes are dropped automatically when the constraint itself is dropped.

Support for more complex foreign key relationships

The following topics show you how to use foreign keys to enforce more complex relationships between tables.

Multiple columns

Foreign keys can reference multiple columns. The list of columns form a key that corresponds to a table's primary key or to a backing index. The referencing table contains foreign keys of the referenced table key.

In the following example, the enforced foreign key definitions indicate that:

Each SongName value in the TopHits table must have a matching value in the Songs table.
Each SingerFirstName and SingerLastName pair of values must have a matching FirstName and LastNamepair of values in the Singers table.

GoogleSQL

CREATE TABLE TopHits (
  Rank INT64 NOT NULL,
  SongName STRING(MAX),
  SingerFirstName STRING(MAX),
  SingerLastName STRING(MAX),

  -- Song names must either be NULL or have matching values in Songs.
  FOREIGN KEY (SongName) REFERENCES Songs (SongName),

  -- Singer names must either be NULL or have matching values in Singers.
  FOREIGN KEY (SingerFirstName, SingerLastName)
  REFERENCES Singers (FirstName, LastName)

) PRIMARY KEY (Rank);

PostgreSQL

CREATE TABLE TopHits (
  Rank BIGINT NOT NULL,
  SongName VARCHAR,
  SingerFirstName VARCHAR,
  SingerLastName VARCHAR,

  -- Song names must either be NULL or have matching values in Songs.
  FOREIGN KEY (SongName) REFERENCES Songs (SongName),

  -- Singer names must either be NULL or have matching values in Singers.
  FOREIGN KEY (SingerFirstName, SingerLastName)
  REFERENCES Singers (FirstName, LastName),

  PRIMARY KEY (Rank)
);

Circular references

Occasionally tables have circular dependencies, perhaps for legacy reasons or due to denormalization. Spanner foreign keys permit circular references. Since a referenced table must exist before a foreign key can reference it, one of the foreign keys must be added with an ALTER TABLE statement. Here's an example

Create TableA, without a foreign key.
Create TableB with a foreign key constraint on TableA.
Use ALTER TABLE on TableA to create a foreign key reference to TableB.

Self-referencing tables

One special type of circular reference is a table that defines a foreign key that references the same table. For example, the following snippet shows a foreign key to enforce that an employee's ManagerId is also an employee.

GoogleSQL

CREATE TABLE Employees (
  EmployeeId INT64 NOT NULL,
  EmployeeName STRING(MAX) NOT NULL,
  ManagerId INT64,
  FOREIGN KEY (ManagerId) REFERENCES Employees (EmployeeId)
) PRIMARY KEY (EmployeeId);

PostgreSQL

CREATE TABLE Employees (
  EmployeeId BIGINT NOT NULL,
  EmployeeName VARCHAR NOT NULL,
  ManagerId BIGINT,
  FOREIGN KEY (ManagerId) REFERENCES Employees (EmployeeId),
  PRIMARY KEY (EmployeeId)
);

What's next

Learn more about foreign keys support in Spanner.
Learn more about Spanner's INFORMATION SCHEMA.