Last update: 5 October, 2016: added examples of using sp_getapplock and sp_releaseapplock.
Introduction
I ran into this problem lately. I had a stored procedure that did some global stuff for a database, it would do the exact same thing every time you executed it. It had no parameters.
It would at the best case be useless and use unnecessary server resources when executed multiple times concurrently, and in the worst case would cause deadlocks in that situation.
And this is exactly what this post is about: a stored procedure that is not supposed to be executed multiple times concurrently.
Examples are stored procedures that are purging a database log, updating derived tables or columns, etc.
Problem
You can start a stored procedure, and before it is finished, can start it again from a different connection.
Solutions
There are (at least) two possible solutions for this problem.
- Use sp_getapplock and sp_releaseapplock (the preferred solution). My credits go to Andy Novick who has published about this before on MSSQLTips. I have added examples of using ‘Session’ as lock owner, and therefore I hope that my article will still have added value.
- My initial solution using dynamic management views is still in this article for reference.
Using sp_getapplock and sp_releaseapplock
How it works
For your convenience the links to MSDN for help on those two system stored procedures:
MSDN page about sp_getapplock
MSDN page about sp_releaseapplock
Important to understand is that as @LockOwner you can either use a transaction or a session.
I have included two examples for both types of lockowner.
Using ‘Transaction’ as lock owner
In this stored procedure example (of which you can download the script) sp_getapplock is used with @LockOwner = ‘Transaction’.
When you use this example, it is really important to use the TRY .. CATCH code so that the lock is always released before the stored procedure execution ends.
-------------- [sp].[StoredProcUsingAppLock] --------------
IF EXISTS (SELECT * FROM sys.objects WHERE object_id = OBJECT_ID(N'[sp].[StoredProcUsingAppLock]') AND type in (N'P', N'PC'))
DROP PROCEDURE [sp].[StoredProcUsingAppLock]
GO
SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
/*
==========================================================================================
Author: Hans Michiels
Create date: 5-oct-2016
Description: Example of stored procedure that is using sp_getapplock by using a transaction.
==========================================================================================
*/
CREATE PROCEDURE [sp].[StoredProcUsingAppLock]
(
@RaiseError BIT = 0
)
AS
BEGIN
SET NOCOUNT ON;
DECLARE @RC INT
DECLARE @message VARCHAR(500)
BEGIN TRY
--\
---) Protection Against Concurrent Executions
--/
BEGIN TRAN
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': Try to obtain a lock ..'
RAISERROR(@message, 0, 1) WITH NOWAIT;
EXEC @RC = sp_getapplock
@Resource = 'StoredProcUsingAppLock',
@LockMode = 'Exclusive',
@LockOwner = 'Transaction',
@LockTimeout = 60000 -- 1 minute
IF @RC < 0
BEGIN
IF @@TRANCOUNT > 0 ROLLBACK TRAN;
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': Sorry, could not obtain a lock within the timeout period, return code was ' + CONVERT(VARCHAR(30), @RC) + '.'
RAISERROR(@message, 0, 1) WITH NOWAIT;
RETURN @RC
END ELSE BEGIN
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': AppLock obtained ..'
RAISERROR(@message, 0, 1) WITH NOWAIT;
END
--\
---) Stored procedure body
--/
-- Wait so that stored procedure has some considerable execution time.
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': Hello World!'
RAISERROR(@message, 0, 1) WITH NOWAIT;
DECLARE @i INT = 0
WHILE @i < 6
BEGIN
WAITFOR DELAY '00:00:01';
EXEC [sp].[SubStoredProc];
SET @i = @i + 1
END
IF @RaiseError = 1
BEGIN
RAISERROR('An error on demand', 16, 1);
END
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': Goodbye ..'
RAISERROR(@message, 0, 1) WITH NOWAIT;
COMMIT TRAN
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': Transaction committed, appLock released ..'
RAISERROR(@message, 0, 1) WITH NOWAIT;
END TRY
BEGIN CATCH
IF @@TRANCOUNT > 0 ROLLBACK TRAN;
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': Transaction rolled back, appLock released ..'
RAISERROR(@message, 0, 1) WITH NOWAIT;
THROW;
END CATCH
END
GO
When executed ..
I would not be me if I did not include a test and a proof that it works.
As you can see in the printscreens below, all executions as nicely after each other ..
As you can see in the printscreens below, this even works well, when an error occurs, due to the TRY .. CATCH block ..
Using ‘Session’ as lock owner
In this stored procedure example (of which you can download the script) sp_getapplock is used with @LockOwner = ‘Session’.
An “advantage” of using the session might be that you can also use it, when you do not want to use a transaction within your stored procedure.
As with the previous example, now also it is really important to use the TRY .. CATCH code so that the lock is always released before the stored procedure execution ends.
IF EXISTS (SELECT * FROM sys.objects WHERE object_id = OBJECT_ID(N'[sp].[StoredProcUsingAppLockNoTran]') AND type in (N'P', N'PC'))
DROP PROCEDURE [sp].[StoredProcUsingAppLockNoTran]
GO
SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
/*
==========================================================================================
Author: Hans Michiels
Create date: 5-oct-2016
Description: Example of stored procedure that is using sp_getapplock without using a transaction.
==========================================================================================
*/
CREATE PROCEDURE [sp].[StoredProcUsingAppLockNoTran]
(
@RaiseError BIT = 0
)
AS
BEGIN
SET NOCOUNT ON;
DECLARE @RC INT
DECLARE @message VARCHAR(500)
BEGIN TRY
--\
---) Protection Against Concurrent Executions
--/
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': Try to obtain a lock ..'
RAISERROR(@message, 0, 1) WITH NOWAIT;
EXEC @RC = sp_getapplock
@Resource = 'StoredProcUsingAppLockNoTran',
@LockMode = 'Exclusive',
@LockOwner = 'Session',
@LockTimeout = 60000 -- 1 minute
IF @RC < 0
BEGIN
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': Sorry, could not obtain a lock within the timeout period, return code was ' + CONVERT(VARCHAR(30), @RC) + '.'
RAISERROR(@message, 0, 1) WITH NOWAIT;
RETURN @RC
END ELSE BEGIN
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': AppLock obtained ..'
RAISERROR(@message, 0, 1) WITH NOWAIT;
END
--\
---) Stored procedure body
--/
-- Wait so that stored procedure has some considerable execution time.
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': Hello World!'
RAISERROR(@message, 0, 1) WITH NOWAIT;
DECLARE @i INT = 0
WHILE @i < 6
BEGIN
WAITFOR DELAY '00:00:01';
EXEC [sp].[SubStoredProc];
SET @i = @i + 1
END
IF @RaiseError = 1
BEGIN
RAISERROR('An error on demand', 16, 1);
END
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': Goodbye ..'
RAISERROR(@message, 0, 1) WITH NOWAIT;
EXEC @RC = sp_releaseapplock @Resource='StoredProcUsingAppLockNoTran', @LockOwner='Session';
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': AppLock released ..'
RAISERROR(@message, 0, 1) WITH NOWAIT;
END TRY
BEGIN CATCH
EXEC @RC = sp_releaseapplock @Resource='StoredProcUsingAppLockNoTran', @LockOwner='Session';
SELECT @message = CONVERT(VARCHAR(30), GETDATE(), 121) + ': AppLock released after error ..'
RAISERROR(@message, 0, 1) WITH NOWAIT;
THROW;
END CATCH
END
GO
When executed ..
As you can see in the printscreens below, all executions as nicely after each other ..
As you can see in the printscreens below, this even works well, when an error occurs, due to the TRY .. CATCH block ..
Other considerations
As you might have noticed, this only worked because the timeout was 60 seconds, and all the executions could be completed within those 60 seconds. If this would not be the case, the stored procedure execution that was waiting to obtain the applock would time out and would bail out due to a RETURN statement.
So you might play a bit with the timeout time to influence this behavior.
If you do not care that the non-first executions bail out, you could set the time out to a smaller value, e.g. 5000 (milliseconds).
If you absolutely want all executions to continue after waiting, you could set the time out to a higher value, that is above the maximum execution time of the stored procedure.
In this way you can choose to let non-first executions wait or not, as could be achieved by using the @WaitIfExecuted parameter in my initial solution.
My initial solution using dynamic management views
So if you want to create a stored procedure that is Protected Against Concurrent Executions, you have to do it yourself.
Making a PACE stored procedure is not very difficult. There is however one big pitfall. The examples on the internet I found all have nasty bug in it, at least when your stored procedure executes other stored procedures: when this happens, the “main” stored procedure goes off the radar, when the dynamic management objects sys.dm_exec_requests and sys.dm_exec_sql_text() are used, that means, you cannot detect that it is running as long as the “sub” stored procedure is being executed.
To make the challenge even bigger, the object_id of the “main” stored procedure is nowhere, so with the available dynamic management objects there is no way to solve this.
Therefore an extra “user” dynamic management table is needed, I called it [udm_storedproc_executions]. It is automaticly cleaned up in the stored procedure [sp].[GetPace].
This stored procedure is used to implement Protection Against Concurrent Executions.
You execute it at the beginning in stored procedures that you want to protect against concurrent executions.
How it works
- You create the table [dbo].[udm_storedproc_executions] (or name it differently) in your database.
- You create the stored procedure [sp].[GetPace] (or name it differently) in your database.
- You add the code snippet below to stored procedures that are not supposed to run multiple times concurrently.
Create the table
(c) Copyright 2016 - hansmichiels.com
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see
*/
IF OBJECT_ID('[dbo].[udm_storedproc_executions]', 'U') IS NOT NULL
DROP TABLE [dbo].[udm_storedproc_executions];
GO
SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
SET ANSI_PADDING OFF
GO
CREATE TABLE [dbo].[udm_storedproc_executions](
[start_time] [datetime] NOT NULL,
[session_id] [smallint] NOT NULL,
[connection_id] [uniqueidentifier] NOT NULL,
[task_address] [varbinary](8) NOT NULL,
[object_id] [int] NOT NULL,
CONSTRAINT [PK_dbo_udm_storedproc_executions] PRIMARY KEY CLUSTERED
( [start_time] ASC,
[session_id] ASC,
[connection_id] ASC,
[task_address] ASC
) WITH (PAD_INDEX = OFF, STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF, ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
) ON [PRIMARY]
GO
SET ANSI_PADDING OFF
GO
Create the stored procedure
(c) Copyright 2016 - hansmichiels.com
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see
*/
-------------- [sp].[GetPace] --------------
IF EXISTS (SELECT * FROM sys.objects WHERE object_id = OBJECT_ID(N'[sp].[GetPace]') AND type in (N'P', N'PC'))
DROP PROCEDURE [sp].[GetPace]
GO
SET ANSI_NULLS ON
GO
SET QUOTED_IDENTIFIER ON
GO
/*
==========================================================================================
Author: Hans Michiels
Create date: 17-sep-2016
Description: Generic stored procedure that can give a different stored procedure
PACE (Protection Against Concurrent Execution).
==========================================================================================
*/
CREATE PROCEDURE [sp].[GetPace]
(
@StoredProcedureName NVARCHAR(256), -- Quoted name including schema, e.g. [dbo].[SomeStoredProcedure]
@WaitIfExecuted BIT = 0, -- Indication whether to wait when the stored procedure is already executing (and execute it afterwards).
@PaceAction VARCHAR(8) OUTPUT, -- Domain values [ CONTINUE | WAIT | RETURN ]
@InsertDmRow BIT = 0, -- Indication if a row must be inserted to table [dbo].[udm_storedproc_executions]
@EnableDebugMessages BIT = 0 -- Speaks for itself
)
AS
BEGIN
SET NOCOUNT ON;
--\
---) Declarations.
--/
DECLARE @SpidThatWins INT
DECLARE @WaitTime CHAR(8) = '00:00:03'
DECLARE @message NVARCHAR(500)
--\
---) Insert new row into table [dbo].[udm_storedproc_executions].
--/
IF @InsertDmRow = 1
BEGIN
INSERT INTO [dbo].[udm_storedproc_executions]
SELECT
der.[start_time],
der.[session_id],
der.[connection_id],
der.[task_address],
OBJECT_ID(@StoredProcedureName, 'P') AS [object_id]
FROM
sys.dm_exec_requests der
CROSS APPLY
sys.dm_exec_sql_text(der.sql_handle) xst
WHERE
der.session_id = @@SPID
AND der.[sql_handle] is not null
AND xst.objectid = OBJECT_ID('[sp].[GetPace]', 'P');
END
-- We need to prevent that this stored procedure would be executed multiple times
-- simultaneously.
SELECT @SpidThatWins = -- The one that first started. If started at exact the same time, the lowest SPID wins.
(
SELECT TOP 1 der.[session_id]
FROM
sys.dm_exec_requests der
JOIN
dbo.udm_storedproc_executions spx
ON spx.[session_id] = der.[session_id]
AND spx.[start_time] = der.[start_time]
AND spx.[connection_id] = der.[connection_id]
AND spx.[task_address] = der.[task_address]
WHERE
der.[sql_handle] is not null
AND spx.[object_id] = OBJECT_ID(@StoredProcedureName, 'P')
ORDER BY
der.[start_time] ASC,
der.[session_id] ASC
)
SELECT @PaceAction =
CASE
WHEN @@SPID = @SpidThatWins THEN 'CONTINUE'
WHEN @WaitIfExecuted = 0 THEN 'RETURN'
WHEN @WaitIfExecuted = 1 THEN 'WAIT'
END
IF @PaceAction = 'WAIT'
BEGIN
-- Still executing, so wait until I can start.
IF @EnableDebugMessages = 1
BEGIN
-- Still executing, so wait until I can start.
SET @message = @StoredProcedureName + N' is waiting for other execution to finish .. ';
RAISERROR(@message, 0, 1) WITH NOWAIT;
END
WAITFOR DELAY @WaitTime
END
IF @PaceAction = 'RETURN' AND @EnableDebugMessages = 1
BEGIN
-- Still executing, so wait until I can start.
SET @message = N'Cancelling execution of stored proc ' + @StoredProcedureName;
RAISERROR(@message, 0, 1) WITH NOWAIT;
END
--\
---) Cleanup old rows from [udm_storedproc_executions].
--/
DELETE FROM spx
FROM
[dbo].[udm_storedproc_executions] spx
LEFT JOIN
sys.dm_exec_requests der
ON spx.[session_id] = der.[session_id]
AND spx.[start_time] = der.[start_time]
AND spx.[connection_id] = der.[connection_id]
AND spx.[task_address] = der.[task_address]
WHERE
der.session_id IS NULL;
END
GO
Add the code snippet
AS
BEGIN
SET NOCOUNT ON;
--\
---) PACE (Protection Against Concurrent Executions) code
--/
DECLARE @PaceAction VARCHAR(8), @InsertDmRow BIT = 1
WHILE 1 = 1
BEGIN
EXECUTE [sp].[GetPace]
@StoredProcedureName = '[sp].[StoredProcGettingPace]',
@WaitIfExecuted = 1,
@PaceAction = @PaceAction OUTPUT,
@InsertDmRow = @InsertDmRow,
@EnableDebugMessages = 1;
IF @PaceAction = 'RETURN' RETURN 0;
IF @PaceAction = 'CONTINUE' BREAK;
SET @InsertDmRow = 0;
END
--\
---) Stored procedure body ..
--/
END
GO
Proof that it works
For the demo I use the following stored procedures:
- [sp].[SubStoredProc]
- [sp].[StoredProcGettingPace]
- [sp].[StoredProcGettingPaceNoWait]
By the way, all the scripts can be downloaded here.
[sp].[SubStoredProc] is being executed by both other stored procedures.
[sp].[StoredProcGettingPace]
First I will demonstrate the testresults with [sp].[StoredProcGettingPace].
It is executed from 4 query windows at the same time. Debug messages show us what happened.
The query that predicts the order in which the session_ids will be processed: the order is 60, 59, 57, 52.
As expected session 60 is processed first.
As expected session 59 is processed after 60. The “Hello world” time (the beginning of the actual stored procedure body) is after the “Goodbye” time of session 60.
Then session 57. Again the “Hello world” time is after the “Goodbye” time of session 59.
And finally session 52. It goes without saying that the “Hello world” time is after the “Goodbye” time of the previous session.
[sp].[StoredProcGettingPaceNoWait]
It is important to understand parameter @WaitIfExecuted of stored procedure [sp].[GetPace].
In the previous example we used the value 1 (true), which means that every execution of a stored procedure protected with [sp].[GetPace] is eventually executed, but later.
But if we use the value 0 (false) for @WaitIfExecuted, we actually cancel the execution of a stored procedure if it is already running.
Here are the testresults of executing [sp].[StoredProcGettingPaceNoWait].
Again it is executed from 4 query windows at the same time and debug messages show us what happened.
As expected the stored procedure is executed only once. The other executions are cancelled (see below).
Download all the scripts here.
Conclusion / Wrap up
In this blog post you could read about two different solutions that can prevent a stored procedure from being executed multiple times simultaneously.
This can be convenient in certain situations, when the stored procedure does some “global” work in your database, and it would be unnecessary and potentially “dangerous” when it is executing multiple times concurrently.
(c) 2016 hansmichiels.com – Do not steal the contents – spread the link instead – thank you.
A simpler approach would be to create a shared temporary table if it does not exist (##TableName), and then select * from the temporary table and wrap the entire procedure innards in a transaction with the transaction isolation level switched to serializable for the duration of the stored proc. That should do the same thing as all connections would be attempting to access the same shared temporary table, the serializable isolation should hold the lock for the duration of the transaction (stored proc innards) preventing the rest of the proc being executed by 2nd+ concurrent connections.
Hi Ian, thanks for the suggestion. This could be a valid approach as well. Your approach could however have a bigger impact on existing stored procedures and would require more testing and rework. You would not only need a transaction (what if multiple not really related updates, inserts or deleted are done and everything is in one big transaction, you might get more locking than desired) but also try catch code or similar, to ensure that the global temporary table is deleted before exiting the stored procedure. My approach has less impact on existing stored procedures: their behavior will not change, only the part at the beginning of the stored procedure is added. But there will be use cases for both ways. Thanks for yours.
Wouldn’t using the built in sp_getapplock functionally be a simpler approach?
Thank you all for the useful suggestions.
Especially the use of sp_getapplock is something I overlooked.
This proves a few things:
– I am human and not perfect
– You can learn from blogging, usually when writing and investigating, but now after publication.
Wouldn’t it be simpler to use the built in sp_getapplock functionality? Here’s a sample implementation:
https://www.mssqltips.com/sqlservertip/3202/prevent-multiple-users-from-running-the-same-sql-server-stored-procedure-at-the-same-time/
Hi Hans. This is what app locks are for. You can call sp_getapplock at the beginning of the SP and sp_releaseapplock at the end to achieve the same result more transparently
Thank you all for the useful suggestions.
Especially the use of sp_getapplock is something I overlooked.
This proves a few things:
– I am human and not perfect
– You can learn from blogging, usually when writing and investigating, but now after publication.
I once gave this same problem some thought but never had to implement it. In software engineering, the solution to the problem you described is referred to as a Singleton, a process that will only have one instance of itself executing at a time.
If you want a block to ensure exclusive execution while a single stored procedure is executing, I suggest doing 2 things:
1. Create an empty table with one column, “id”, in some database that can be accessed by any process executing on the server.
2. In the stored procedure, the first instruction should be:
“SELECT id FROM Table WITH (TABLOCKX);”. This will create an exclusive lock on the empty table and will cause every other process attempting to access the empty table to wait.
When the executing stored procedure finishes executing, the exclusive table lock will automatically be dropped allowing other processes to execute the stored procedure, one at a time.
Thanks Lee for your reply. Both the sp_getapplock and your approach are valid ways to deal with this problem.
Hello. In your @LockOwner = ‘Transaction’ example, the calls to sp_releaseapplock are missing.
Hello Alexandre,
Thank you for letting me know, I will update it when I can.
Hi Hans,
Is it possible to consume the messages from the RAISERROR on and log the message through an SSIS package without having to make any changes to the Stored Procedures?
Thanks,
Shashi
Hi Shashi, I am not sure what you mean. If an error is unhandled in a stored procedure, it can be caught in a SSIS Package, based on the settings in the package.
If you catch an error in a stored procedure, and then use THROW or RAISERROR to raise it again, this is also the case, but some of the original error information (like the number) might get lost.
Hope this was helpful, if not, please explain what your use case is. Thanks.
Best regards,
Hans