yahoo-eng-team team mailing list archive

[Sean Dague <sean@xxxxxxxxx>]

Probably needs to come in as spec as this requires work outside of the
virt driver

** Tags added: vmware

** Summary changed:

- [sr-iov] pci passthrough whitelist doesn't support mult node
+ [vmware] [sr-iov] pci passthrough whitelist doesn't support mult node

** Changed in: nova
       Status: New => Opinion

** Changed in: nova
   Importance: Undecided => Wishlist

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to OpenStack Compute (nova).
https://bugs.launchpad.net/bugs/1612029

Title:
  [vmware] [sr-iov] pci passthrough whitelist doesn't support mult node

Status in OpenStack Compute (nova):
  Opinion

Bug description:
  Description
  ===========
  pci_passthrough_whitelist = {"address": "0000:0a:00.1", "physical_network": "physnet1"}
  If a nova-compute supports multi nodes. The nodes's pci address maybe repeat.
  So format above can not describe the situation.

  Openstack architecture is supporting mapping a nova-compute to multi node.
  For drivers listed in Openstack community code, there is only vmware driver supporting it .
  But there are many other hypervisor products which based on KVM/Xen/PowerVM. They do enhancement On KVM/Xen/PowerVM and provide managebility and develop their drivers supporting mapping a nova-compute to multi node.
  When customers want to build a cloud and make a decision on how to choose cloud products and hypervisor products, they maybe choose hypervisor beyond KVM/Xen/PowerVM. So openstack cloud products should regard sr-iov working with multi-node.


  Steps to reproduce
  ==================
  1.Configure nova-compute mappint to multi node.
  2.there are two node whic have same pci_address
  3.pci_passthrough_whitelist can not describe the situation

  Expected result
  ===============
  None

  Suggest Solution
  =============
  pci_passthrough_whitelist = {"address": "0000:0a:00.1", "node": "xxxx", "physical_network": "physnet1"}
  we add node description in pci_passthrough_whitelist.
  when nova-compute process pci_passthrough_whitelist per node, ResourceTracker and PciDeviceStats just process own node's pci_passthrough_whitelist.
  PciDeviceSpec doesn't need to add "Node" item. We will pop "Node" of pci_passthrough_whitelist.
  The "Node" is for selection per node.

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/1612029/+subscriptions