Using Bound Sessions
Bound sessions ease the coordination of actions across multiple sessions on the same server. Bound sessions allow two or more sessions to share the same transaction and locks, and can work on the same data without lock conflicts. Bound sessions can be created from multiple sessions within the same application or from multiple applications with separate sessions.
To participate in a bound session, a session calls sp_getbindtoken or srv_getbindtoken (through Open Data Services) to get a bind token. A bind token is a character string that uniquely identifies each bound transaction. The bind token is then sent to the other sessions to be bound with the current session. The other sessions bind to the transaction by calling sp_bindsession, using the bind token received from the first session.
|A session must have an active user transaction in order for sp_getbindtoken or srv_getbindtoken to succeed.|
Bind tokens must be transmitted from the application code that makes the first session to the application code that subsequently binds their sessions to the first session. There is no Transact-SQL statement or API function that an application can use to get the bind token for a transaction started by another process. Some of the methods that can be used to transmit a bind token include the following:
If the sessions are all initiated from the same application process, bind tokens can be stored in global memory or passed into functions as a parameter.
If the sessions are made from separate application processes, bind tokens can be transmitted using interprocess communication (IPC), such as a remote procedure call (RPC) or dynamic data exchange (DDE).
Bind tokens can be stored in a table in an instance of the SQL Server Database Engine that can be read by processes wanting to bind to the first session.
Only one session in a set of bound sessions can be active at any time. If one session is executing a statement on the instance or has results pending from the instance, no other session bound to it can access the instance until the current session finishes processing or cancels the current statement. If the instance is busy processing a statement from another of the bound sessions, an error occurs indicating that the transaction space is in use and the session should retry later.
When you bind sessions, each session retains its isolation level setting. Using SET TRANSACTION ISOLATION LEVEL to change the isolation level setting of one session does not affect the setting of any other session bound to it.
The two types of bound sessions are local and distributed.
Local bound session
Allows bound sessions to share the transaction space of a single transaction in a single instance of the Database Engine.
Distributed bound session
Allows bound sessions to share the same transaction across two or more instances until the entire transaction is either committed or rolled back by using Microsoft Distributed Transaction Coordinator (MS DTC).
Distributed bound sessions are not identified by a character string bind token; they are identified by distributed transaction identification numbers. If a bound session is involved in a local transaction and executes an RPC on a remote server with SET REMOTE_PROC_TRANSACTIONS ON, the local bound transaction is automatically promoted to a distributed bound transaction by MS DTC and an MS DTC session is started.
In earlier versions of SQL Server, bound sessions were primarily used in developing extended stored procedures that must execute Transact-SQL statements on behalf of the process that calls them. Having the calling process pass in a bind token as one parameter of the extended stored procedure allows the procedure to join the transaction space of the calling process, thereby integrating the extended stored procedure with the calling process.
In the SQL Server 2005 Database Engine, stored procedures written using CLR are more secure, scalable, and stable than extended stored procedures. CLR-stored procedures use the SqlContext object to join the context of the calling session, not sp_bindsession.
Bound sessions can be used to develop three-tier applications in which business logic is incorporated into separate programs that work cooperatively on a single business transaction. These programs must be coded to carefully coordinate their access to a database. Because the two sessions share the same locks, the two programs must not try to modify the same data at the same time. At any point in time, only one session can be doing work as part of the transaction; there can be no parallel execution. The transaction can only be switched between sessions at well-defined yield points, such as when all DML statements have completed and their results have been retrieved.