yahoo-eng-team team mailing list archive

[OpenStack Infra <1889633@xxxxxxxxxxxxxxxxxx>]

Reviewed:  https://review.opendev.org/744020
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=737e0c0111acd364d1481bdabd9d23bc8d5d6a2e
Submitter: Zuul
Branch:    master

commit 737e0c0111acd364d1481bdabd9d23bc8d5d6a2e
Author: Stephen Finucane <stephenfin@xxxxxxxxxx>
Date:   Thu Jul 30 17:37:38 2020 +0100

    tests: Add reproducer for bug #1889633
    
    With the introduction of the cpu-resources work [1], (libvirt) hosts can
    now report 'PCPU' inventory separate from 'VCPU' inventory, which is
    consumed by instances with pinned CPUs ('hw:cpu_policy=dedicated'). As
    part of that effort, we had to drop support for the ability to boot
    instances with 'hw:cpu_thread_policy=isolate' (i.e. I don't want
    hyperthreads) on hosts with hyperthreading. This had been previously
    implemented by marking thread siblings of the host cores used by such an
    instance as reserved and unusable by other instances, but such a design
    wasn't possible in world where we had to track resource consumption in
    placement before landing in the host. Instead, the 'isolate' policy now
    simply means "give me a host without hyperthreads". This is enforced by
    hosts with hyperthreads reporting the 'HW_CPU_HYPERTHREADING' trait, and
    instances with the 'isolate' policy requesting
    'HW_CPU_HYPERTHREADING=forbidden'.
    
    Or at least, that's how it should work. We also have a fallback query
    for placement to find hosts with 'VCPU' inventory and that doesn't care
    about the 'HW_CPU_HYPERTHREADING' trait. This was envisioned to ensure
    hosts with old style configuration ('[DEFAULT] vcpu_pin_set') could
    continue to be scheduled to. We figured that this second fallback query
    could accidentally pick up hosts with new-style configuration, but we
    are also tracking the available and used cores from those listed in the
    '[compute] cpu_dedicated_set' as part of the host 'NUMATopology' objects
    (specifically, via the 'pcpuset' and 'cpu_pinning' fields of the
    'NUMACell' child objects). These are validated by both the
    'NUMATopologyFilter' and the virt driver itself, which means hosts with
    new style configuration that got caught up in this second query would be
    rejected by this filter or by a late failure on the host. (Hint: there's
    much more detail on this in the spec).
    
    Unfortunately we didn't think about hyperthreading. If a host gets
    picked up in the second request, it might well have enough PCPU
    inventory but simply be rejected in the first query since it had
    hyperthreads. In this case, because it has enough free cores available
    for pinning, neither the filter nor the virt driver will reject the
    request, resulting in a situation whereby the instance ends up falling
    back to the old code paths and consuming $flavor.vcpu host cores, plus
    the thread siblings for each of these cores. Despite this, it will be
    marked as consuming $flavor.vcpu VCPU (not PCPU) inventory in placement.
    
    This patch proves this to be the case, allowing us to resolve the issue
    later.
    
    [1] https://specs.openstack.org/openstack/nova-specs/specs/train/approved/cpu-resources.html
    
    Change-Id: I87cd4d14192b1a40cbdca6e3af0f818f2cab613e
    Signed-off-by: Stephen Finucane <stephenfin@xxxxxxxxxx>
    Related-Bug: #1889633


** Changed in: nova
       Status: In Progress => Fix Released

-- 
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/1889633

Title:
  Pinned instance with thread policy can consume VCPU

Status in OpenStack Compute (nova):
  Fix Released
Status in OpenStack Compute (nova) train series:
  Triaged
Status in OpenStack Compute (nova) ussuri series:
  Triaged

Bug description:
  In Train, we introduced the concept of the 'PCPU' resource type to
  track pinned instance CPU usage. The '[compute] cpu_dedicated_set' is
  used to indicate which host cores should be used by pinned instances
  and, once this config option was set, nova would start reporting
  'PCPU' resource types in addition to (or entirely instead of, if
  'cpu_shared_set' was unset) 'VCPU'. Requests for pinned instances (via
  the 'hw:cpu_policy=dedicated' flavor extra spec or equivalent image
  metadata property) would result in a query for 'PCPU' inventory rather
  than 'VCPU', as previously done.

  We anticipated some upgrade issues with this change, whereby there
  could be a period during an upgrade in which some hosts would have the
  new configuration, meaning they'd be reporting PCPU, but the remainder
  would still be on legacy config and therefore would continue reporting
  just VCPU. An instance could be reasonably expected to land on any
  host, but since only the hosts with the new configuration were
  reporting 'PCPU' inventory and the 'hw:cpu_policy=dedicated' extra
  spec was resulting in a request for 'PCPU', the hosts with legacy
  configuration would never be consumed.

  We worked around this issue by adding support for a fallback placement
  query, enabled by default, which would make a second request using
  'VCPU' inventory instead of 'PCPU'. The idea behind this was that the
  hosts with 'PCPU' inventory would be preferred, meaning we'd only try
  the 'VCPU' allocation if the preferred path failed. Crucially, we
  anticipated that if a host with new style configuration was picked up
  by this second 'VCPU' query, an instance would never actually be able
  to build there. This is because the new-style configuration would be
  reflected in the 'numa_topology' blob of the 'ComputeNode' object,
  specifically via the 'cpuset' (for cores allocated to 'VCPU') and
  'pcpuset' (for cores allocated to 'PCPU') fields. With new-style
  configuration, both of these are set to unique values. If the
  scheduler had determined that there wasn't enough 'PCPU' inventory
  available for the instance, that would implicitly mean there weren't
  enough of the cores listed in the 'pcpuset' field still available.

  Turns out there's a gap in this thinking: thread policies. The
  'isolate' CPU thread policy previously meant "give me a host with no
  hyperthreads, else a host with hyperthreads but mark the thread
  siblings of the cores used by the instance as reserved". This didn't
  translate to a new 'PCPU' world where we needed to know how many cores
  we were consuming up front before landing on the host. To work around
  this, we removed support for the latter case and instead relied on a
  trait, 'HW_CPU_HYPERTHEADING', to indicate whether a host had
  hyperthread support or not. Using the 'isolate' policy meant that
  trait could not be defined on the host, or the trait was "forbidden".
  The gap comes via a combination of this trait request and the fallback
  query. If we request the isolate thread policy, hosts with new-style
  configuration and sufficient PCPU inventory would nonetheless be
  rejected if they reported the 'HW_CPU_HYPERTHEADING' trait. However,
  these could get picked up in the fallback query and the instance would
  not fail to build on the host because of lack of 'PCPU' inventory.
  This means we end up with a pinned instance on a host using new-style
  configuration that is consuming 'VCPU' inventory. Boo.

  # Steps to reproduce

  1. Using a host with hyperthreading support enabled, configure both
  '[compute] cpu_dedicated_set' and '[compute] cpu_shared_set'

  2. Boot an instance with the 'hw:cpu_thread_policy=isolate' extra
  spec.

  # Expected result

  Instance should not boot since the host has hyperthreads.

  # Actual result

  Instance boots.

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/1889633/+subscriptions