openstack restricted to clusters?

[jogersh]

In VOVA all OpenStack VMs had to be in a separate cluster from the regular vSphere VMs (virtual machines)

Is this still the case for VIO,GA version? or can a vSphere cluster contain a mixture of vSphere VMs and VIO instances?

[admin]

Clusters used by OpenStack should *not* be used to create VMs from other mechanisms. This will throw-off the resource accounting done by Nova.

You need to create clusters dedicated for OpenStack tenants. In the same vCenter,  however, you can have other clusters that are not allocated to OpenStack and use them to create non-OpenStack VMs.

[sirishkrvmtn]

There are other OpenStack solutions that do not have this restriction. Specifically, I work at Platform9; and Platform9 Managed OpenStack is more liberal in this regard:

1. vSphere cluster can contain a mixture of vSphere VMs and instances deployed via OpenStack

2. There are no restrictions on what operations can be performed via OpenStack vs VMware vSphere

[admin]

Lets say you have a cluster with cumulative capacity of 24vCPUs. You added this cluster to Nova. Nova sees this as a giant hypervisor of 24vCPU capacity. You create a some VMs using Nova and say they consumed 10 vCPUs. Nova accounting says 14 available. Now if you create VMs on this cluster through some other mechanism and those consume 10vCPUs. Then unless you have a mechanism to constantly communicate resource capacity and usage, you will break the Nova accounting.

For vSphere Operations, lets say you go to vCenter and delete a VM created by Nova. Unless you communicate such actions back to Nova, it will not know that this VM was deleted. Moreoever this VM probably has a Neutron record, a Cinder record in case of disks etc. So all those records and services need to be updated.

Love to learn more on how Platform9 manages the above situations.

[sirishkrvmtn]

Arvind,

Good example. As you mentioned, the answer isn't a one-time reconciliation between vSphere accounting and OpenStack services; but an ongoing set of updates that reconcile these. And that's what our discovery services do within Platform9. For context, this is the architecture that we were proposing within VMware vCloud Director as well back in 2012; because customers complained about being unable to perform certain operations via vSphere once vCloud Director was in use.

So, in your example, Platform9 Managed OpenStack will correctly detect that 10 vCPUs have been consumed by a new workload <x> on the concerned resource.

This isn't just necessary for vSphere interoperability; it is also necessary because in the real world, there can always be changes to system state outside of the changes initiated by Nova. If you think about the management fabric running with Amazon Web Services, Google Compute Engine; etc; they accept and embrace the eventual-consistency model.

Best,

- Sirish.