Primary Key patterns for federated (horizontally partitioned) databases
https://stackoverflow.com/questions/11338064
italiano
english
français
española
中国
日本の
العربية
Deutsch
한국어
Português
RussianQuestion
I apologize in advance if this is a duplicate, I'm not sure exactly what I should be searching for.
Recently I watched this nifty video on Channel 9 about database federation in SQL Azure. At first I was surprised by the fact that identity columns are not supported by SQL Azure Federation, but it makes sense. If you have a table that is split between 2 (or more) databases, and they have no way to share their identity increment value, you could end up with 2 (or more) entities sharing the same primary key.
The video touched lightly on how to deal with this problem, mentioning using something like a Guid for a PK instead of an integer-based identity column. I know that at least MSSQL will create a clustered index on the PK by default, and it is bad to have a clustered index on a Guid (or uniqueidentifier) type. I'm by no means a relational expert, but I also believe that Guid's have a performance degradation when compared to integer-based types for PK-based lookups.
So this got me wondering, what might an integer-based PK pattern look like for a horizontally federated database look like? Back when I wrote old school EJB's and we needed to generate an integer for a primary key, we had to have a separate sequence lookup table to replace the rdbms identity/auto-increment features. I recall this was was painful, as many other things with EJB's were back then.
What are the commonly-accepted patterns, if any, for assigning primary keys in a horizontally federated database table? Given my current knowledge I would lean away from Guid's because of performance degradation, and I would lean away from sequential lookup indices because of development cost and added complexity (mostly development cost). What should I be leaning toward?
Solution
Something like this might be useful, depending on the nature of your app
