Comment 17 for bug 1180044

cut-and-paste of comments posted on review. I want to be sure to preserve the knowledge I gained from working this bug for several weeks. By all means, Gary please finish the work. Thank you.

Conceptually, all VMs in a vCenter are placed under a Datacenter and each datacenter has a vmfolder that holds all instances. The idea of a datacenter (at least conceptually) is that this represents a segregated hunk of hardware so when you ask about all virtual machines in a datacenter in Cleveland, it makes no sense to talk about the ones in Munich (as an exaggerated example nobody really does that).

So selecting the correct datacenter paths is pretty important. vmfolder happens to be an implicit and hidden path for VM placement specification even so, Nova must tell vCenter "the scheduler said place it here" where here is the vmfolder, datastore, etc. Since the current scheduler is only giving the datastore coordinate we can extrapolate which datacenter the scheduler meant by calculating the nearest in tree vmfolder path.
Currently the driver completely ignores all this and we tell people to just use one datacenter so that the calculation can't possibly get messed up (but more than one and the path calculation can be wrong) so this is a pretty major driver change we need to get right.

We need to be sure that we encode the tree traversals correctly so that the scheduler's intentions are properly translated into vCenter commands. That's not happening right now.
I will pull this for manual testing later this week but I can tell by looking you'll still fail when there are more than one datacenter and the intended datacenter is not *first* or position 0.