yahoo-eng-team team mailing list archive

[Matt Riedemann <mriedem.os@xxxxxxxxx>]

In this case the compute service is managing baremetal nodes using the
ironic driver, correct? In that case yes the ComputeFilter will reject
all nodes managed by that compute service host because it has been
disabled, just like it would in a 1:1 host:node (hypervisor) topology
when using the libvirt driver for example.

One question is why are you disabling the hypervisor if you still want
things scheduled to it? If you're doing an upgrade / maintenance on the
compute service host, then you likely don't want new scheduling requests
to go to it because they could fail while the nova-compute service is
down.

This doesn't really sound like a bug since it is the designed behavior.
At most it's an opinion but you haven't really stated a use case besides
"I want to be able to schedule baremetal instances on nodes managed by a
disabled service" which is probably not something we're going to
support.

There is the concept of the ironic compute service hashring to have HA
management of the same ironic nodes, maybe that's something you should
be investigating?

https://specs.openstack.org/openstack/nova-
specs/specs/newton/implemented/ironic-multiple-compute-hosts.html

https://docs.openstack.org/ironic/stein/install/configure-
compute.html?highlight=hash%20ring

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

-- 
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/1831195

Title:
  disable one compute service will prevent scheduler  to choose the
  hypervisor the compute service manange

Status in OpenStack Compute (nova):
  Invalid

Bug description:
  Description
     ===========
    When one nova compute service manage multi hypervisors, disable the nova-compute service will prevent nova scheduler from choosing all the hypervisor it manages because the code in compute_filter.py :

    service = host_state.service
    if service['disabled']:
     LOG.debug("%(host_state)s is disabled, reason: %(reason)s",
         {'host_state': host_state,
          'reason': service.get('disabled_reason')})
     return False

    and i think this is not reasonable.

    One solution is to make each compute service manage only one
  hypervisor, but this result more compute services.

     Steps to reproduce
     ==================

     Expected result
     ===============
     operator can disable individual hypervisor

     Actual result
     =============
     Disable one nova compute prevent nova scheduer choosing all the hypervisors it manages

     Environment
     ===========

     Logs & Configs
     ==============

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/1831195/+subscriptions