Open Stack

On 04/08/2014 11:27 AM, Mike Spreitzer wrote:

> There really should be one more step in that flow.  Consider a create
> scenario.  In general, as the client makes the calls to create
> individual resources: some will succeed, some will fail (some in ways
> that make it clear the capacity will not be used, others not), for some
> the client will issue the request but timeout waiting for success, and
> for some the client will never even issue the call to create.  After the
> client has done as much as it will, there should be another call back to
> the holistic scheduling function to release the capacity that was
> planned to be used but the client has given up on.

While this is probably a good idea, it's also probably not sufficient. 
I suspect we'll also want to have those initial reservations time out in 
some relatively short period of time (capped at the max amount of time 
we expect the instantiation phase to take) if they are not confirmed.

Chris