Composite Keys and Referential Integrity in T-SQL - sql

Question

Is it possible, in T-SQL, to have a relationship table with a composite key composed of 1 column defining Table Type and another column defining the Id of a row from a table referenced in the Table Type column?
For a shared-email address example:Three different user tables (UserA, UserB, UserC)One UserType Table (UserType)One Email Table (EmailAddress)One Email-User Relationship Table (EmailRelationship)The EmailRelationship Table contains three columns, EmailId, UserTypeId and UserId
Can I have a relationship from each User table to the EmailRelationship table (or some other way?) to maintain referential integrity?
I've tried making all three columns in the EmailRelationship table into primary keys, I've tried making only UserTypeId and UserId primary.
CREATE TABLE [dbo].[UserType](
[Id] [int] IDENTITY(1,1) NOT NULL ,
[Type] [varchar](50) NOT NULL)
insert into [dbo].[UserType]
([Type])
values
('A'),('B'),('C')
CREATE TABLE [dbo].[UserA](
[Id] [int] IDENTITY(1,1) NOT NULL,
[UserTypeId] [int] NOT NULL,
[Name] [varchar](50) NOT NULL)
insert into [dbo].[UserA]
(UserTypeId,Name)
values
(1,'UserA')
CREATE TABLE [dbo].[UserB](
[Id] [int] IDENTITY(1,1) NOT NULL,
[UserTypeId] [int] NOT NULL,
[Name] [varchar](50) NOT NULL)
insert into [dbo].[UserB]
(UserTypeId,Name)
values
(2,'UserB')
CREATE TABLE [dbo].[UserC](
[Id] [int] IDENTITY(1,1) NOT NULL,
[UserTypeId] [int] NOT NULL,
[Name] [varchar](50) NOT NULL)
insert into [dbo].[UserC]
(UserTypeId,Name)
values
(3,'UserC')
CREATE TABLE [dbo].[Email](
[Id] [int] IDENTITY(1,1) NOT NULL,
[EmailAddress] [varchar](50) NOT NULL)
insert into [dbo].[email]
(EmailAddress)
values
('SharedEmail#SharedEmail.com')
CREATE TABLE [dbo].[EmailRelationship](
[EmailId] [int] NOT NULL,
[UserTypeId] [int] NOT NULL,
[UserId] [int] NOT NULL)
insert into [dbo].[EmailRelationship]
(EmailId, UserTypeId, UserId)
values
(1,1,1),(1,2,1),(1,3,1)

Answer 1

No there isn't, a foreign key can refer to one table, and one table only, I can think of three ways you could approach this.
The first is to have 3 columns, one for each user table, each column with a foreign key, and a check constraint to check that at one, and only one of the values is not null
CREATE TABLE dbo.EmailRelationship
(
EmailId INT NOT NULL,
UserTypeId INT NOT NULL,
UserAId INT NULL,
UserBId INT NULL,
UserCId INT NULL,
CONSTRAINT FK_EmailRelationship__UserAID FOREIGN KEY (UserAId)
REFERENCES dbo.UserA (Id),
CONSTRAINT FK_EmailRelationship__UserBID FOREIGN KEY (UserBId)
REFERENCES dbo.UserB (Id),
CONSTRAINT FK_EmailRelationship__UserCID FOREIGN KEY (UserCId)
REFERENCES dbo.UserC (Id),
CONSTRAINT CK_EmailRelationship__ValidUserId CHECK
(CASE WHEN UserTypeID = 1 AND UserAId IS NOT NULL AND ISNULL(UserBId, UserCId) IS NULL THEN 1
WHEN UserTypeID = 2 AND UserBId IS NOT NULL AND ISNULL(UserAId, UserCId) IS NULL THEN 1
WHEN UserTypeID = 3 AND UserCId IS NOT NULL AND ISNULL(UserAId, UserBId) IS NULL THEN 1
ELSE 0
END = 1)
);
Then as a quick example trying to insert a UserAId with a user Type ID of 2 gives you an error:
INSERT EmailRelationship (EmailID, UserTypeID, UserAId)
VALUES (1, 1, 1);
The INSERT statement conflicted with the CHECK constraint "CK_EmailRelationship__ValidUserId".
The second approach is to just have a single user table, and store user type against it, along with any other common attributes
CREATE TABLE dbo.[User]
(
Id INT IDENTITY(1, 1) NOT NULL,
UserTypeID INT NOT NULL,
Name VARCHAR(50) NOT NULL,
CONSTRAINT PK_User__UserID PRIMARY KEY (Id),
CONSTRAINT FK_User__UserTypeID FOREIGN KEY (UserTypeID) REFERENCES dbo.UserType (UserTypeID),
CONSTRAINT UQ_User__Id_UserTypeID UNIQUE (Id, UserTypeID)
);
-- NOTE THE UNIQUE CONSTRAINT, THIS WILL BE USED LATER
Then you can just use a normal foreign key constraint on your email relationship table:
CREATE TABLE dbo.EmailRelationship
(
EmailId INT NOT NULL,
UserId INT NOT NULL,
CONSTRAINT PK_EmailRelationship PRIMARY KEY (EmailID),
CONSTRAINT FK_EmailRelationship__EmailId
FOREIGN KEY (EmailID) REFERENCES dbo.Email (Id),
CONSTRAINT FK_EmailRelationship__UserId
FOREIGN KEY (UserId) REFERENCES dbo.[User] (Id)
);
It is then no longer necessary to store UserTypeId against the email relationship because you can join back to User to get this.
Then, if for whatever reason you do need specific tables for different user types (this is not unheard of), you can create these tables, and enforce referential integrity to the user table:
CREATE TABLE dbo.UserA
(
UserID INT NOT NULL,
UserTypeID AS 1 PERSISTED,
SomeOtherCol VARCHAR(50),
CONSTRAINT PK_UserA__UserID PRIMARY KEY (UserID),
CONSTRAINT FK_UserA__UserID_UserTypeID FOREIGN KEY (UserID, UserTypeID)
REFERENCES dbo.[User] (Id, UserTypeID)
);
The foreign key from UserID and the computed column UserTypeID back to the User table, ensures that you can only enter users in this table where the UserTypeID is 1.
A third option is just to have a separate junction table for each User table:
CREATE TABLE dbo.UserAEmailRelationship
(
EmailId INT NOT NULL,
UserAId INT NOT NULL,
CONSTRAINT PK_UserAEmailRelationship PRIMARY KEY (EmailId, UserAId),
CONSTRAINT FK_UserAEmailRelationship__EmailId FOREIGN KEY (EmailId)
REFERENCES dbo.Email (Id),
CONSTRAINT FK_UserAEmailRelationship__UserAId FOREIGN KEY (UserAId)
REFERENCES dbo.UserA (Id)
);
CREATE TABLE dbo.UserBEmailRelationship
(
EmailId INT NOT NULL,
UserBId INT NOT NULL,
CONSTRAINT PK_UserBEmailRelationship PRIMARY KEY (EmailId, UserBId),
CONSTRAINT FK_UserBEmailRelationship__EmailId FOREIGN KEY (EmailId)
REFERENCES dbo.Email (Id),
CONSTRAINT FK_UserBEmailRelationship__UserBId FOREIGN KEY (UserBId)
REFERENCES dbo.UserB (Id)
);
Each approach has it's merits and drawbacks, so you would need to assess what is best for your scenario.

Answer 2

No it does not work that way. You cannot use a column value as a dynamic reference to different tables.
In general the data design is flawed.

Answer 3

Thanks to #GarethD I created a CHECK constraint that called a scalar-function that would enforce referential integrity (only upon insert, refer to caveat below):
Using my above example:
alter FUNCTION [dbo].[UserTableConstraint](#Id int, #UserTypeId int)
RETURNS int
AS
BEGIN
IF EXISTS (SELECT Id From [dbo].[UserA] WHERE Id = #Id and UserTypeId = #UserTypeId)
return 1
ELSE IF EXISTS (SELECT Id From [dbo].[UserB] WHERE Id = #Id and UserTypeId = #UserTypeId)
return 1
ELSE IF EXISTS (SELECT Id From [dbo].[UserC] WHERE Id = #Id and UserTypeId = #UserTypeId)
return 1
return 0
end;
alter table [dbo].[emailrelationship]
--drop constraint CK_UserType
with CHECK add constraint CK_UserType
CHECK([dbo].[UserTableConstraint](UserId,UserTypeId) = 1)
I am sure there is a not insignificant overhead to a Scalar-function call from within a CONSTRAINT. If the above becomes prohibitive I will report back here, though the tables in question will not have to deal with a large volume of INSERTs.
If there are any other reasons to not do the above, I would like to hear them. Thanks!
Update:
I've tested INSERT and UPDATE with 100k rows (SQL Server 2014, 2.1ghz quadcore w/ 8gb ram):
INSERT takes 2 seconds with out the CONSTRAINT
and 3 seconds with the CHECK CONSTRAINT
Turning on IO and TIME STATISTICS causes the INSERT tests to run in:
1.7 seconds with out the CONSTRAINT
and 10 seconds with the CHECK CONSTRAINT
I left the STATISTICS on for the UPDATE 100k rows test:
just over 1sec with out the CONSTRAINT
and 1.5sec with the CHECK CONSTRAINT
My referenced tables (UserA, UserB, UserC from my example) only contain around 10k rows each, so anybody else looking to implement the above may want to run some additional testing, especially if your referenced tables contain millions of rows.
Caveat:
The above solution may not be suitable for most uses, as the only time referential integrity is checked is during the CHECK CONSTRAINT upon INSERT. Any other operations or modifications of the data needs to take that into account. For example, using the above, if an Email is deleted any related EmailRelationship entries will be pointing to invalid data.