This error occurs when you use a try/catch block inside of a transaction. Let's consider a trivial example:
SET XACT_ABORT ON
IF object_id('tempdb..#t') IS NOT NULL
DROP TABLE #t
CREATE TABLE #t (i INT NOT NULL PRIMARY KEY)
BEGIN TRAN
INSERT INTO #t (i) VALUES (1)
INSERT INTO #t (i) VALUES (2)
INSERT INTO #t (i) VALUES (3)
INSERT INTO #t (i) VALUES (1) -- dup key error, XACT_ABORT kills the batch
INSERT INTO #t (i) VALUES (4)
COMMIT TRAN
SELECT * FROM #t
When the fourth insert causes an error, the batch is terminated and the transaction rolls back. No surprises so far.
Now let's attempt to handle that error with a TRY/CATCH block:
SET XACT_ABORT ON
IF object_id('tempdb..#t') IS NOT NULL
DROP TABLE #t
CREATE TABLE #t (i INT NOT NULL PRIMARY KEY)
BEGIN TRAN
INSERT INTO #t (i) VALUES (1)
INSERT INTO #t (i) VALUES (2)
BEGIN TRY
INSERT INTO #t (i) VALUES (3)
INSERT INTO #t (i) VALUES (1) -- dup key error
END TRY
BEGIN CATCH
SELECT ERROR_MESSAGE()
END CATCH
INSERT INTO #t (i) VALUES (4)
/* Error the Current Transaction cannot be committed and
cannot support operations that write to the log file. Roll back the transaction. */
COMMIT TRAN
SELECT * FROM #t
We caught the duplicate key error, but otherwise, we're not better off. Our batch still gets terminated, and our transaction still gets rolled back. The reason is actually very simple:
TRY/CATCH blocks don't affect transactions.
Due to having XACT_ABORT ON, the moment the duplicate key error occurs, the transaction is doomed. It's done for. It's been fatally wounded. It's been shot through the heart...and the error's to blame. TRY/CATCH gives SQL Server...a bad name. (sorry, couldn't resist)
In other words, it will NEVER commit and will ALWAYS be rolled back. All a TRY/CATCH block can do is break the fall of the corpse. We can use the XACT_STATE() function to see if our transaction is committable. If it is not, the only option is to roll back the transaction.
SET XACT_ABORT ON -- Try with it OFF as well.
IF object_id('tempdb..#t') IS NOT NULL
DROP TABLE #t
CREATE TABLE #t (i INT NOT NULL PRIMARY KEY)
BEGIN TRAN
INSERT INTO #t (i) VALUES (1)
INSERT INTO #t (i) VALUES (2)
SAVE TRANSACTION Save1
BEGIN TRY
INSERT INTO #t (i) VALUES (3)
INSERT INTO #t (i) VALUES (1) -- dup key error
END TRY
BEGIN CATCH
SELECT ERROR_MESSAGE()
IF XACT_STATE() = -1 -- Transaction is doomed, Rollback everything.
ROLLBACK TRAN
IF XACT_STATE() = 1 --Transaction is commitable, we can rollback to a save point
ROLLBACK TRAN Save1
END CATCH
INSERT INTO #t (i) VALUES (4)
IF @@TRANCOUNT > 0
COMMIT TRAN
SELECT * FROM #t
Triggers always execute within the context of a transaction, so if you can avoid using TRY/CATCH inside them, things are much simpler.
For a solution to your problem, a CLR Stored Proc could connect back to SQL Server in a separate connection to execute the dynamic SQL. You gain the ability to execute the code in a new transaction and the error handling logic is both easy to write and easy to understand in C#.