Naming convention for unique constraint
Naming conventions are important, and primary key and foreign key have commonly used and obvious conventions (PK_Table and FK_Table_ReferencedTable, respectively). The IX_Table_Column naming for indexes is also fairly standard.
What about the UNIQUE constraint? Is there a commonly accepted naming convention for this constraint? I've seen UK_TableName_Column, UQ_TableName_Column, and someone recommending AX_TableName_Column - I don't know where that comes from.
I've typically used UQ but I don't particularly like it, and I do not enjoy having to defend my choice of using it against a UK advocate.
I would simply like to see if there is a consensus on the most prevalent naming, or a good reasoning as to why one makes more sense than the others.
My naming convention for indices and constraints:
| Index/Constraint Type | Naming Convention |
|---|---|
| Primary key |
<table-name>_PK
|
| Unique index/constraint |
<table-name>_AK{xx}
|
| Non-Unique index |
<table-name>_IX{xx}
|
| Check constraint |
<table-name>_CK{xx}
|
| Default constraint |
<table-name>_DF{xx}
|
| Foreign key constraint |
<table-name>_FK{xx}
|
Where {xx} is a 2-digit sequence number, starting at 01 for each constraint type per table. Primary key doesn't get a sequence number since there can be only one. The 2-char alpha suffix meanings are:
| Suffix | Meaning |
|---|---|
PK |
Primary Key |
AK |
Alternate Key |
FK |
Foreign Key |
IX |
IndeX |
CK |
ChecK |
DF |
DeFault |
I generally want to group metadata/system catalog data by the controlling object rather than by object type.
My thinking is it isn't a key: it's a constraint.
It could be used as a key of course, and uniquely identifies a row, but it isn't the key.
An example would be that the key is "ThingID", a surrogate key used in place of ThingName the natural key. You still need to constrain ThingName: it won't be used as a key though.
I'd also use UQ and UQC (if clustered).
You could use a unique index instead and go for "IXU". By the logic employed, an index is also a key but only when unique. Otherwise it's an index. So then we'd start with IK_columnname for unique indexes and IX_columnname for non-unique indexes. Marvellous.
And the only difference between a unique constraint and a unique index is INCLUDE columns.
Edit: Feb 2013. Since SQL Server 2008, indexes can have filters too. Constraints can not
So, it comes down to one of
- stick with UQ as per the rest of the SQL-using planet
- use IK for unique indexes (IKC for clustered too) to be consistent...