I am agreeing with you here - there is a disconnect between what the engine is doing and how the log is interpreting things.
Would it be difficult for us to log:
1) When the log code starts a statement + the sql that caused it (I believe we have some of this already)
Something like - New Transaction ID: NNNNN, SQL: UPDATE 'bb'...
2) When the log code ends / writes a message - "ID: NNNNN ended/written"
3) When the log code is rolling back something - rolling back ID: NNNNN
4) When a query is added to an existing message.
I'll be taking a look at the code myself to see about this, but figured I would ask the expert : )
I'm hoping that this can provide us with additional insight into how the log is managing / interpreting the queries the randgen is throwing at it. It's proving difficult to try to reproduce this any other way without more info on what the server is doing.
I am agreeing with you here - there is a disconnect between what the engine is doing and how the log is interpreting things.
Would it be difficult for us to log:
1) When the log code starts a statement + the sql that caused it (I believe we have some of this already)
Something like - New Transaction ID: NNNNN, SQL: UPDATE 'bb'...
2) When the log code ends / writes a message - "ID: NNNNN ended/written"
3) When the log code is rolling back something - rolling back ID: NNNNN
4) When a query is added to an existing message.
I'll be taking a look at the code myself to see about this, but figured I would ask the expert : )
I'm hoping that this can provide us with additional insight into how the log is managing / interpreting the queries the randgen is throwing at it. It's proving difficult to try to reproduce this any other way without more info on what the server is doing.