Atomic block behaviour

[pkaczynski]

This would be a question actually, but it will clarify the documentation for me, so I would be grateful if you could elaborate on this. What would be the assumed behaviour in the following scenario:

with transaction.atomic():
    # Code A
    with transaction.atomic():
        x = get_next_value('test')
        # Code B
    # Code C

1. Thread is invoking Code B instructions - are other threads blocked waiting (on which line?) until it ends to run Code B? What about Code A?
2. Thread is invoking Code C instructions. Are other threads blocked waiting (on which line?) until it ends to run Code C? Code B? Code A?

I know the library guarantees than get_next_value always return concurrent gapless values. But for example if I assign this value to an Invoice number in Code B and something goes wrong, there may be a chance that the next transaction would get the next number (in case execution of Code B does not block other threads).

Thank in advance for any answer.

[aaugustin]

If two threads run this code concurrently, thread 2 will block on the get_next_value call and won't start executing # Code B until thread 1 finishes executing # Code C.

This is fairly easy to test by replacing # Code X with print("enter X"); time.sleep(10); print("exit X") and running the script in two terminals side by side.

If you try it, I'm interested in a confirmation that the behavior is indeed what I predicted.

PS - if you enabled the ATOMIC_REQUESTS option (which you probably did if you care enough about transactions to look at django-sequences) then the outer atomic block is created by Django and wraps the entire view.

[pkaczynski]

Thanks for the reply. I can confirm your elaboration. I tested it "manually" using debugger and indeed, thread 2 would block until Code C is fully executed. It works like nesting atomic blocks.

And as to your PS - I did not enable ATOMIC_REQUESTS, and did it deliberately. I do not want requests to be atomic. This is actually the only case I need locking and syncing between requests. I do not want to loose performance :]

Thanks again for the clarification!