I'm in the process of designing a fairly complex system. One of our primary concerns is supporting SQL Server peer-to-peer replication. The idea is to support several geographically separated nodes.
A secondary concern has been using a modern ORM in the middle tier. Our first choice has always been Entity Framework, mainly because the developers like to work with it. (They love the LiNQ support.)
So here's the problem:
With peer-to-peer replication in mind, I settled on using uniqueidentifier with a default value of newsequentialid() for the primary key of every table. This seemed to provide a good balance between avoiding key collisions and reducing index fragmentation.
However, it turns out that the current version of Entity Framework has a very strange limitation: if an entity's key column is a uniqueidentifier (GUID) then it cannot be configured to use the default value (newsequentialid()) provided by the database. The application layer must generate the GUID and populate the key value.
So here's the debate:
- abandon Entity Framework and use another ORM:
- use NHibernate and give up LiNQ support
- use linq2sql and give up future support (not to mention get bound to SQL Server on DB)
- abandon GUIDs and go with another PK strategy
- devise a method to generate sequential GUIDs (COMBs?) at the application layer
I'm leaning towards option 1 with linq2sql (my developers really like linq2[stuff]) and 3. That's mainly because I'm somewhat ignorant of alternate key strategies that support the replication scheme we're aiming for while also keeping things sane from a developer's perspective.
Any insight or opinion would be greatly appreciated.
Another option (not available when this was posted) is to upgrade to EF 4, which supports server-generated GUIDs.
I second Craig's suggestion - option 4.
You can always use the GUID column, populated by the middle-tier, as your PRIMARY KEY (that's a LOGICAL construct).
To avoid massive index (thus: table) fragmentation, use some other key (ideally an INT IDENTITY column) as the CLUSTERING KEY - that's a physical database construct, which CAN be separated from the primary key.
By default, the primary key is the clustering key - but that doesn't have to be that way. In fact, I improved performance and drastically lowered fragmentation by doing just that on a database I "inherited" - add a INT IDENTITY column and put the clustering key on that small, ever-increasing, never-changing INT - works like a charm!
Marc
Huh? I think your three options are a false choice. Consider option 4:
4) Use the Entity Framework with non-sequential, client-generated GUIDs.
The EF can't see DB-server-generated GUIDs for new rows inserted by the framework itself, sure, but you don't need to generate the GUIDs on the DB server. You can generate them on the client when you create your entity instances. The whole point of a GUID is it doesn't matter where you generate it. As for GUIDs generated by a replicated DB, the EF will see them just fine.
Your client-side GUIDs won't be sequential (use Guid.NewGuid()), but they will be world-wide, guaranteed unique.
We do this in shipping, production software with replication. It does work.
Why not use identity column? If you are doing merge replication you can have each system start at a separate seed and work in one direction (e.g. node a starts at 1 and adds 1, node b starts at 0 and subtracts one)...
You can use stored procedures if you are really stuck on using NewSequentialID(). You can bind the result columns from the procedure to the appropriate property and once inserted the SQL-generated GUID will be fed back into the object.
Unfortunately you have to define SPs for all three operations (insert, update, delete) even though the other operations would complete properly using the defaults. You also need to maintain the SP code and ensure it is synchronized with your EF model as you make changes, which may make this option unattractive on account of the additional overhead.
There is a step-by-step example at http://blogs.msdn.com/bags/archive/2009/03/12/entity-framework-modeling-action-stored-procedures.aspx which is pretty straight-forward.
use newseqid with your own orm (it not that hard) with linq
来源：https://stackoverflow.com/questions/604273/advice-please-sql-server-identity-vs-unique-identifier-keys-when-using-entity-f