There are two tables:
- User
- Address
User contains a reference to Address.
Address contains the columns CreatedBy and ModifiedBy, which is reference to User.
How do I design this database to avoid a cyclic dependency?
4
-
4Are you sure the ModifiedBy isn't a reference to the database user that made the change, rather than the applications user (which would reside in the User table)? Either way, it doesn't really matter. I don't see why there's an actual problem? – Philᵀᴹ May 6 '13 at 11:01
-
Which one will created first. As CreatedBy ANd Modifiedby is a must data. And user table should not have addressid as null.. Chicken-egg problem. As i solve this by creating a new table UserAddress which contains the reference for userid and addressid – Shashi May 6 '13 at 11:07
-
2You could always used deferred constraints if your DBMS supports them. – Colin 't Hart May 6 '13 at 11:08
-
NB The relational model supports doing two inserts or updates as one atomic operation, it's a real shame this isn't supported in SQL (as deferred constraints -- even though I suggested them -- are quite horrible). – Colin 't Hart May 7 '13 at 16:50
Add a comment
|
Instead of searching for tips and tricks (deferred constraints included) I would suggest that you simply design your way out of this "reference lock" -- so try something like this:
Facts
- User
(UserID)exists. - Address
(AddressID)was created by User(UserID). - Address
(AddressID)was created on Date(DateCreated). - Address
(AddressID)was last modified by User(UserID)on Date(ModifiedOn). - User
(UserID)resides at Address(AddressID)since Date(ValidFrom).
Constraints
EachAddress was created byexactly oneUser.It is possible that more than oneAddress was created bythe sameUser.EachAddress was created onexactly oneDate.It is possible that more than oneAddress was created onthe sameDate.For eachAddressandDate,thatAddress was modified byat most oneUser onthatDate.For eachUserandDate,thatUser residesat most oneAddress sincethatDate.
Logical
As far as mandatory address is concerned, verify that on the application layer and wrap the loading statements into a transaction -- that way you'll get all or nothing.
You don't have a choice but to create the cyclic dependency in 2 operations as below because one table does not exist when you create the first one.
CREATE TABLE A (A_ID INT PRIMARY KEY, B_FK INT);
CREATE TABLE B (B_ID INT PRIMARY KEY, A_FK INT REFERENCES A(A_ID));
ALTER TABLE A ADD B_FK INT;
If you wish to avoid cyclic dependency. Then you need to remove one REFERENCES constraint or you can add a DELETE and UPDATE CASCADE reference in one way. You could also implement a TRIGGER if your logic is somewhat complex.
-
1Removing the constraints removes the cyclic dependency from the definition, but not the design. You could add an Events table to record the UserID and AddressID that last created or modified the address, but that just moves the dependency one step further away. Going the other way, if the User table had CreatedBy and ModifiedBy columns, the cyclic dependency would exist in one table. This is similar to an employee table with a supervisor column where the supervisor is also an employee. As Phil indicated - Not a problem. – Leigh Riffel May 6 '13 at 17:32
-
@LeighRiffel I agree. But the Events table you propose does actually remove all cyclic dependencies. – ypercubeᵀᴹ May 6 '13 at 18:18
-
@ypercube Indeed it does; not sure how I got that wire crossed. To be clear you probably shouldn't create an Events table even though it does remove the cyclic dependency. – Leigh Riffel May 6 '13 at 18:26
-
Anyway, I don't think this answer addresses the problem. The question (I think) is about how to avoid the cyclic path at all, not how to create the FKs with the cyclic path in the first place. – ypercubeᵀᴹ May 6 '13 at 19:52
