yahoo-eng-team team mailing list archive

[melanie witt <2088831@xxxxxxxxxxxxxxxxxx>]

Public bug reported:

I discovered this issue while working on adding flavor scanning to
'nova-manage limits migrate_to_unified_limits' [1].

When unified limits is enabled ([quota]driver =
nova.quota.UnifiedLimitsDriver), the quota checking code calls
nova.scheduler.utils.resources_for_limits(flavor, is_bfv) to obtain a
list of resource classes for which quota limits need to be enforced.

This check however fails when the following exception is raised and the
API returns a 400 status code:

  {"badRequest": {"code": 400, "message": "Memory encryption requested
by hw:mem_encryption extra spec in zqdxonddgkblwxarzpjj flavor but image
None doesn't have 'hw_firmware_type' property set to 'uefi' or volume-
backed instance was requested"}}

This is due to reuse of the
nova.virt.hardware.get_mem_encryption_constraint(flavor, image) method
in nova.scheduler.utils. In any case where the image metadata is not
properly populated in the request spec passed to
ResourceRequest.from_request_spec(request_spec),
nova.virt.hardware.get_mem_encryption_constraint(flavor, image) will
raise a FlavorImageConflict exception and cause things to fail.

Example from nova/scheduler/utils.py [2] where a "fake" RequestSpec is
used:

def _get_resources(flavor, is_bfv):
    # create a fake RequestSpec as a wrapper to the caller
    req_spec = objects.RequestSpec(flavor=flavor, is_bfv=is_bfv)

    # TODO(efried): This method is currently only used from places that
    # assume the compute node is the only resource provider.  So for now, we
    # just merge together all the resources specified in the flavor and pass
    # them along.  This will need to be adjusted when nested and/or shared RPs
    # are in play.
    res_req = ResourceRequest.from_request_spec(req_spec)

    return res_req.merged_resources()

This bug is similar to https://bugs.launchpad.net/nova/+bug/2040449 and
https://bugs.launchpad.net/nova/+bug/2007697.


[1] https://review.opendev.org/c/openstack/nova/+/924110
[2] https://github.com/openstack/nova/blob/1acaf899a6964484e5b5be4337618ebbe6ca8dbb/nova/scheduler/utils.py#L653-L664

** Affects: nova
     Importance: Undecided
         Status: New


** Tags: quotas

-- 
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/2088831

Title:
  Server create with hw:mem_encryption fails with unified limits quotas

Status in OpenStack Compute (nova):
  New

Bug description:
  I discovered this issue while working on adding flavor scanning to
  'nova-manage limits migrate_to_unified_limits' [1].

  When unified limits is enabled ([quota]driver =
  nova.quota.UnifiedLimitsDriver), the quota checking code calls
  nova.scheduler.utils.resources_for_limits(flavor, is_bfv) to obtain a
  list of resource classes for which quota limits need to be enforced.

  This check however fails when the following exception is raised and
  the API returns a 400 status code:

    {"badRequest": {"code": 400, "message": "Memory encryption requested
  by hw:mem_encryption extra spec in zqdxonddgkblwxarzpjj flavor but
  image None doesn't have 'hw_firmware_type' property set to 'uefi' or
  volume-backed instance was requested"}}

  This is due to reuse of the
  nova.virt.hardware.get_mem_encryption_constraint(flavor, image) method
  in nova.scheduler.utils. In any case where the image metadata is not
  properly populated in the request spec passed to
  ResourceRequest.from_request_spec(request_spec),
  nova.virt.hardware.get_mem_encryption_constraint(flavor, image) will
  raise a FlavorImageConflict exception and cause things to fail.

  Example from nova/scheduler/utils.py [2] where a "fake" RequestSpec is
  used:

  def _get_resources(flavor, is_bfv):
      # create a fake RequestSpec as a wrapper to the caller
      req_spec = objects.RequestSpec(flavor=flavor, is_bfv=is_bfv)

      # TODO(efried): This method is currently only used from places that
      # assume the compute node is the only resource provider.  So for now, we
      # just merge together all the resources specified in the flavor and pass
      # them along.  This will need to be adjusted when nested and/or shared RPs
      # are in play.
      res_req = ResourceRequest.from_request_spec(req_spec)

      return res_req.merged_resources()

  This bug is similar to https://bugs.launchpad.net/nova/+bug/2040449
  and https://bugs.launchpad.net/nova/+bug/2007697.

  
  [1] https://review.opendev.org/c/openstack/nova/+/924110
  [2] https://github.com/openstack/nova/blob/1acaf899a6964484e5b5be4337618ebbe6ca8dbb/nova/scheduler/utils.py#L653-L664

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/2088831/+subscriptions