© 2004
Canonical Ltd.
•
Terms of use
•
Data privacy
•
Contact Launchpad Support
•
Blog
•
Careers
•
System status
•
0e1f616
(Get the code!)
I think we are dealing with more than one bug here.
It looks like there is an issue around the ROLLBACK TO SAVEPOINT code. If we remove all SAVEPOINT references from the test, failures occur MUCH less frequently.
Without SAVEPOINTS, there is still the rare failure. In this case, it looks like the master is having a failed statement rolled back, but the the TransactionServices code is failing to recognize it for some reason and does not rollback the GPB message appropriately. This one is more difficult to reproduce, but in the cases that I've seen, it has always been on a failing statment when AUTOCOMMIT=ON.