跨负载平衡应用程序的并发和锁定

Question

I am writing an application where users can create items with a start date and an end date and save them to a SQL database hosted in Microsoft Sql Server. The rule in the application is that only a single item can be active for a given time (no overlapping items). The application also needs to be load balanced, which means (as far as I know) traditional semaphores / locking won't work.

A few additional items:

• The records are persisted into two tables (based on a business requirement).
• Users are allowed to "insert" records in the middle of an existing record. Inserted records adjust the start & end dates of any pre-existing records to prevent overlapping items (if necessary).
• Optimally we want to accomplish this using our ORM and .Net. We don't have as much leeway to make database schema changes but we can create transactions and do other kinds of SQL operations through our ORM.

Our goal is to prevent the following from happening:

• Saves from multiple users resulting in overlapping items in either table (ex. users 1 & 2 query the database, see that there aren't overlapping records, and save at the same time)
• Saves from multiple users resulting in a different state in each of the destination tables (ex. Two users "insert" records, and the action is interleaved between the two tables. Table A looks as though User 1 went first, and table B looks as though User 2 went first.)

My question is how could I lock or prevent multiple users from saving / inserting at the same time across load balanced servers.

Note: We are currently looking into using sp_getapplock as it seems like it would do what we want, if you have experience with this or feel like it would be a bad decision and want to elaborate that would be appreciated as well!

Edit: added additional info

Answer 1

There are at least a couple of options:

• You can create a stored procedure which wraps the INSERT operation in a transaction:

   begin tran select to see if there is an existing record you can: - invalidate the previous record - insert a new record or - raise an error commit tran catch error rollback tran 

• You can employ a last-in-wins strategy where you don't employ a write level lock, but rather a read-level pseuodo-lock, essentially ignoring all records except the latest one.