Uploaded image for project: 'Red Hat OpenStack Services on OpenShift'
  1. Red Hat OpenStack Services on OpenShift
  2. OSPRH-10744

Disable instance_info networking cache interval by default

XMLWordPrintable

    • Default heal_instance_info_cache_interval=0
    • False
    • Hide

      None

      Show
      None
    • False
    • Committed
    • Committed
    • In Progress
    • RHOSSTRAT-51 - Compute Engineering Backlog
    • nova-operator-container-1.0.7-6
    • Committed
    • Committed
    • 0% To Do, 0% In Progress, 100% Done
    • Hide
      .Periodic healing of Nova internal instance info cache is disabled
      By default, `heal_instance_info_cache_interval` is now disabled to improve performance of the neutron API server by removing the load created by the nova-compute agent. This will not affect cache correctness, because it is updated during most VM operations.
      Show
      .Periodic healing of Nova internal instance info cache is disabled By default, `heal_instance_info_cache_interval` is now disabled to improve performance of the neutron API server by removing the load created by the nova-compute agent. This will not affect cache correctness, because it is updated during most VM operations.
    • Enhancement
    • Done
    • Important

      nova-compute can send up to ~9 queries / minute / server to neutron with the default setting of 60. As we look at scaling to ever greater numbers of computes in a cluster (the scale team has gone up to 1000), this can put a significant load on neutron (in the case of 1000 computes, up to 150 req/s).

      In previous discussions with the nova team, it was agreed that setting this to zero in the modern era should not be a problem. So I think we should consider making this the default. If not, then documenting configuration / tuning recommendations for this setting, especially since this tends to affect bigger installs the most.

              sbauza@redhat.com Sylvain Bauza
              twilson@redhat.com Terry Wilson
              rhos-workloads-compute
              Votes:
              0 Vote for this issue
              Watchers:
              6 Start watching this issue

                Created:
                Updated:
                Resolved: