I am using SQL Server 2008 and when I run this Statement in Management studio the Select statement in the Catch Block is executed as expected
BEGIN TRY
INSE
I don't get the CATCH block hit at all.
That's because the code won't compile, because the object doesn't exist, no plan is generated, so nothing runs to hit the CATCH block.
You can never hit this catch block so somethign is wrong with your testing/example. You can hit an outer catch block in a different scope (eg nested stored procs)
Edit: I'm using SQL Server 2005 SP3
It depends when deferred name resolution applies, related to statement level recompilation.
In my case, the whole batch fails both times and no statement level recompilation happens so no deferred name resolution
In OP's case, the batch compiles and runs but then has a statement level recompilation/deferred name resolution error in running code
I'm off to find some references about why it's different, given BOL doesn't say much, neither does Erland Sommarskog
Now that we have all the explanations as to why this is happening. Let's see an actual solution to the problem.
First let's take the statements that @d-k-mulligan proposed above and turn them into stored procs.
IF OBJECT_ID('dbo.prcIDontExistINSERT', 'P') IS NOT NULL DROP PROCEDURE dbo.prcIDontExistINSERT
GO
CREATE PROCEDURE dbo.prcIDontExistINSERT
AS
BEGIN TRY
INSERT INTO IDontExist(ProductID)
VALUES(1)
END TRY
BEGIN CATCH
SELECT 'There was an error! ' + ERROR_MESSAGE()
END CATCH
GO
IF OBJECT_ID('dbo.prcIDontExistSELECT', 'P') IS NOT NULL DROP PROCEDURE dbo.prcIDontExistSELECT
GO
CREATE PROCEDURE dbo.prcIDontExistSELECT
AS
BEGIN TRY
SELECT * FROM IDontExist
END TRY
BEGIN CATCH
SELECT 'There was an error! ' + ERROR_MESSAGE()
END CATCH
GO
If we run either of them we see the same error.
EXEC dbo.prcIDontExistINSERT
EXEC dbo.prcIDontExistSELECT
Msg 208, Level 16, State 1, Procedure prcIDontExistSELECT, Line 4
Invalid object name 'IDontExist'.
The solution now is to create error handling wrapper procs with the sole purpose of catching any error from the original procs above that are getting the object not found errors.
IF OBJECT_ID('dbo.prcIDontExistInsert_ERROR_HANDLER', 'P') IS NOT NULL DROP PROCEDURE dbo.prcIDontExistInsert_ERROR_HANDLER
GO
CREATE PROCEDURE dbo.prcIDontExistInsert_ERROR_HANDLER
AS
BEGIN TRY
EXEC dbo.prcIDontExistINSERT
END TRY
BEGIN CATCH
SELECT 'There was an error! ' + ERROR_MESSAGE()
END CATCH
GO
IF OBJECT_ID('dbo.prcIDontExistSELECT_ERROR_HANDLER', 'P') IS NOT NULL DROP PROCEDURE dbo.prcIDontExistSELECT_ERROR_HANDLER
GO
CREATE PROCEDURE dbo.prcIDontExistSELECT_ERROR_HANDLER
AS
BEGIN TRY
EXEC dbo.prcIDontExistSELECT
END TRY
BEGIN CATCH
SELECT 'There was an error! ' + ERROR_MESSAGE()
END CATCH
GO
Finally, let's run either of our error handling procs and see the message we expect.
EXEC dbo.prcIDontExistInsert_ERROR_HANDLER
EXEC dbo.prcIDontExistSELECT_ERROR_HANDLER
There was an error! Invalid object name 'IDontExist'.
NOTE: Kalman Toth did all the hard research work here: http://www.sqlusa.com/articles2008/trycatch/
Directly from http://msdn.microsoft.com/en-us/library/ms175976.aspx.
USE AdventureWorks2008R2;
GO
BEGIN TRY
-- Table does not exist; object name resolution
-- error not caught.
SELECT * FROM NonexistentTable;
END TRY
BEGIN CATCH
SELECT
ERROR_NUMBER() AS ErrorNumber
,ERROR_MESSAGE() AS ErrorMessage;
END CATCH
The error is not caught and control passes out of the TRY…CATCH construct to the next higher level.
Running the SELECT statement inside a stored procedure will cause the error to occur at a level lower than the TRY block. The error will be handled by the TRY…CATCH construct.
Workaround with dynamic sql. Maybe it will be helpful for someone.
begin try
exec('
insert into IDontExist(ProductID)
values(1)
')
end try
begin catch
select 'There was an error! ' + error_message()
end catch
This behaviour happens if you previously had a table IDontExist
and compiled a plan for it that is still in the cache then drop the table.
It also happens if you run the individual statement twice even without the table ever existing. The first run raises an error that is not caught. The second run (after the first plan is cached) succeeds.
/*Clear Cache*/
DBCC FREEPROCCACHE
GO
BEGIN TRY
INSERT INTO IDontExist(ProductID)
VALUES(1)
END TRY
BEGIN CATCH
SELECT 'There was an error! ' + ERROR_MESSAGE()
END CATCH
GO
/*Plan now Cached*/
SELECT query_plan
FROM sys.dm_exec_cached_plans cp
OUTER APPLY sys.dm_exec_sql_text(plan_handle) t
OUTER APPLY sys.dm_exec_query_plan(plan_handle) qp
WHERE t.text LIKE '%IDontExist%'
OPTION (RECOMPILE)
GO
BEGIN TRY
INSERT INTO IDontExist(ProductID)
VALUES(1)
END TRY
BEGIN CATCH
SELECT 'There was an error! ' + ERROR_MESSAGE()
END CATCH
GO
The INSERT
statement gets auto parameterised.
If you change your Select * from IDontExist
statement to Select * from IDontExist WHERE ProductID = 1
this also becomes auto parameterised and they behave the same.
I'm not absolutely certain why the auto parameterisation makes a difference here. I think that it is explained by the below extract from BOL however.
The following types of errors are not handled by a
CATCH
block when they occur at the same level of execution as theTRY…CATCH
construct ... [those] that occur during statement-level recompilation ... If an error occurs during compilation or statement-level recompilation at a lower execution level (for example, when executingsp_executesql
or a user-defined stored procedure) inside theTRY
block, the error occurs at a lower level than theTRY…CATCH
construct and will be handled by the associatedCATCH
block.
I presume the auto parametrization of that statement means that it gets recompiled at a lower execution level and is catchable.
This has bitten me in the past as well.
Not all errors generated inside the TRY block statements are passed into the CATCH block. Any errors with a severity of 10 or less are considered to be warnings and do not cause control to flow to the CATCH block. Also, any errors that break the database connection will not cause the CATCH block to be reached. There may be other situations as well.