Open Stack

Hi,

I have noticed the following problem, but I'm not sure if it's bug or feature ;)

Scheduler is calculating available ram per node by taking total ram
amount and then subtracting memory for every non-deleted instance
associated to that compute node. That's all fine if all of them are
running. Then the result roughly corresponds with reality.

https://github.com/openstack/nova/blob/stable/essex/nova/scheduler/host_manager.py#L301

But if, say,  50% of instances are stopped most of the time, they are
still accounted for used ram for that compute node. RamFilter will
prevent you to create more instances there..

Is operator supposed to set over-commit in this case ?

It seems that resources are allocated to instance at creation and they
are "locked" during the whole lifecycle. One might expect that stopped
instances would be disassociated with the compute node to free up the
resources. Resources would be then allocated again on next
start-instance, potentially on different compute node. Or is it just
my misunderstanding of cloudiness ?

Regards,

Brano Zarnovican

PS: We are running Essex 2012.1.3