I was testing to see how nested transactions work, and uncovered this disturbing and unexpected behavior.
using(TransactionScope otx = new TransactionScope())
using(SqlConnection conn1 = new SqlConnection("Server=S;Database=DB;Trusted_Connection=yes"))
using(SqlCommand cmd1 = conn1.CreateCommand())
{
conn1.Open();
cmd1.CommandType = CommandType.Text;
cmd1.CommandText = "INSERT INTO FP.ACLs (ChangeToken,ACL) VALUES (1,0x)";
cmd1.ExecuteNonQuery();
using(TransactionScope itx = new TransactionScope(TransactionScopeOption.RequiresNew))
using(SqlConnection conn2 = new SqlConnection("Server=S;Database=DB;Trusted_Connection=yes"))
using(SqlCommand cmd2 = conn1.CreateCommand())
{
conn2.Open();
cmd2.CommandType = CommandType.Text;
cmd2.CommandText = "INSERT INTO FP.ACLs (ChangeToken,ACL) VALUES (2,0x)";
cmd2.ExecuteNonQuery();
// we don't commit the inner transaction
}
otx.Complete(); // nonetheless, the inner transaction gets committed here and two rows appear in the database!
}
I saw this other question, but the solution did not apply.
If I don't specify TransactionScopeOption.RequiresNew (i.e. I don't use a nested transaction, just a nested scope), then the entire transaction is rolled back when the inner scope is not completed, and an error occurs when calling otx.Complete(). This is fine.
But I certainly don't expect a nested transaction to be committed when it did not complete successfully! Does anybody know what is going on here and how I can get the expected behavior?
The database is SQL Server 2008 R2.