yahoo-eng-team team mailing list archive

[Joe Gordon <1190635@xxxxxxxxxxxxxxxxxx>]

This bug sounds like it may be valid to me, by default we allow core
subscription in the scheduler.   So the default settings should allow
over-subscription of CPUs.  That being said,  the bug is fairly vague
and doesn't prove the NoValidHost is coming from a core restriction, so
marking this as incomplete.

** Changed in: nova
       Status: Invalid => Incomplete

-- 
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/1190635

Title:
  default scheduler fails when CPUs are the limiting resource

Status in OpenStack Compute (Nova):
  Incomplete

Bug description:
  http://docs.openstack.org/folsom/openstack-compute/admin/content/ch_scheduling.html
  (likewise, http://docs.openstack.org/grizzly/openstack-compute/admin/content/ch_scheduling.html)

  Indicate that the default scheduler uses:
  scheduler_default_filters=AvailabilityZoneFilter,RamFilter,ComputeFilter
  but no CoreFilter is in the default list.

  Accordingly, if the compute node with the least amount of memory is
  already has all available (possibly overcommitted) vCPUs in use, then
  the scheduler will happily launch the next instance on a node that
  CANNOT run the instance, and nova show will somewhat unclearly report:

  | fault                  | {u'message': u'NoValidHost', u'code': 500,
  u'created': u'2013-06-13T14:40:46Z'}     |

  At the very minimum, NO scheduler should be considered functional if
  it schedules an instance on a node that refuses to run it, while there
  are nodes that would be able to.

  This is running on Ubuntu Precise with Folsom deployed from the cloud
  archive.

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/1190635/+subscriptions