views:

84

answers:

2

I’m interested in learning which technique developers prefer to use to enforce uniqueness in SQL Server: UNIQUE CONSTRAINT or UNIQUE INDEX. Given that there is little difference in the physical implementation of each, how do you decide which is best?

Are there reasons other than performance to evaluate the best solution?

Are there database management advantages to one or the other?

+2  A: 

They are not significantly different. When you create a unique constraint, SQL Server will automatically create a unique index for you.

With the syntax for creating an index, you may have better control defining a unique index to specify clustered/nonclustered, included columns, filegroup, index filtering (SqlSvr2008), etc.

A constraint is preferable in most cases because it expresses the intent of the uniqueness: it is a constraint. An index does not convey this intent.

As for manageability, the impact is minimal. You can manage the index (rebuild, reorg) as if it were created independently of the constraint. The only difference is that the constraint depends on the index, so to drop the index, you must also drop the constraint.

Jeff Meatball Yang
+1 I'd also say consistency, as well as you'll often end up changing a constraint to an index anyway to add an INCLUDE
gbn
+2  A: 

This MSDN article compariing the two is what you're after: http://msdn.microsoft.com/en-us/library/aa224827(SQL.80).aspx

For most purposes, there's no difference - the constraint is implemented as an index under the covers. And though there's the ability to disable the constraint, it doesn't actually work in SQL Server.

It only matters if you want to tweak things like FILLFACTOR, etc for which way you want to implement the unique constraint.

OMG Ponies
The article is good. It's interesting that it concludes "that there's no practical difference" other than meta-data.
bobs
tut, tut quoting MSDN for SQL Server 2000
gbn
@gbn: True, but has anything changed?
OMG Ponies
let's see. INCLUDE = more efficient covering indexes. Filtered indexes = unique constraint over a subset of rows/ignore multiple null etc. These are the 2 headline differences.
gbn