Update statistics from computes if RBD ephemeral is used

[Andrey Grebennikov]

If we use RBD as the backend for ephemeral drives, compute nodes still calculate their available disk size looking back to the local disks.
This is the path how they do it:

* nova/compute/manager.py

    def update_available_resource(self, context):
        """See driver.get_available_resource()

        Periodic process that keeps that the compute host's understanding of
        resource availability and usage in sync with the underlying hypervisor.

        :param context: security context
        """
        new_resource_tracker_dict = {}
        nodenames = set(self.driver.get_available_nodes())
        for nodename in nodenames:
            rt = self._get_resource_tracker(nodename)
            rt.update_available_resource(context)
            new_resource_tracker_dict[nodename] = rt
....................
    def _get_resource_tracker(self, nodename):
        rt = self._resource_tracker_dict.get(nodename)
        if not rt:
            if not self.driver.node_is_available(nodename):
                raise exception.NovaException(
                        _("%s is not a valid node managed by this "
                          "compute host.") % nodename)

            rt = resource_tracker.ResourceTracker(self.host,
                                                  self.driver,
                                                  nodename)
            self._resource_tracker_dict[nodename] = rt
        return rt

* nova/compute/resource_tracker.py

    def update_available_resource(self, context):
        """Override in-memory calculations of compute node resource usage based
        on data audited from the hypervisor layer.

        Add in resource claims in progress to account for operations that have
        declared a need for resources, but not necessarily retrieved them from
        the hypervisor layer yet.
        """
        LOG.audit(_("Auditing locally available compute resources"))
        resources = self.driver.get_available_resource(self.nodename)

* nova/virt/libvirt/driver.py

    def get_local_gb_info():
        """Get local storage info of the compute node in GB.

        :returns: A dict containing:
             :total: How big the overall usable filesystem is (in gigabytes)
             :free: How much space is free (in gigabytes)
             :used: How much space is used (in gigabytes)
        """

        if CONF.libvirt_images_type == 'lvm':
            info = libvirt_utils.get_volume_group_info(
                                 CONF.libvirt_images_volume_group)
        else:
            info = libvirt_utils.get_fs_info(CONF.instances_path)

        for (k, v) in info.iteritems():
            info[k] = v / (1024 ** 3)

        return info

It would be nice to have something like "libvirt_utils.get_rbd_info" which could be used in case CONF.libvirt_images_type == 'rbd'

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/102064

[Dmitry Borodaenko (angdraug)]

Attached the version of the patch for Havana, based on https://github.com/angdraug/nova/tree/rbd-live-migration-stable-havana

[Andrey Grebennikov (agrebennikov)]

It does work for me.

Please check driver.py:
elif CONF.libvirt.images_type == 'rbd':
should be:
elif CONF.libvirt_images_type == 'rbd':

Thanks.

[Dmitry Borodaenko (angdraug)]

Patch for 4.1.x (havana) based on https://github.com/angdraug/nova/tree/rbd-live-migration-stable-havana

[Dmitry Borodaenko (angdraug)]

Patch for 5.x (Icehouse) based on https://github.com/angdraug/nova/tree/rbd-ephemeral-clone-stable-icehouse

[Dmitry Borodaenko (angdraug)]

Icehouse patch version from 6/24 was causing a regression (https://bugs.launchpad.net/mos/+bug/1335628), updated patch attached. Github branch already has the updated version of the patch.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/102064
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=9b559c31b781689fb66551f29a0cb8d10c7bac94
Submitter: Jenkins
Branch: master

commit 9b559c31b781689fb66551f29a0cb8d10c7bac94
Author: Dmitry Borodaenko <email address hidden>
Date: Mon Jun 23 16:12:18 2014 -0700

    Use Ceph cluster stats to report disk info on RBD

    Local disk statistics on compute nodes are irrelevant when ephemeral
    disks are stored in RBD. With RBD, local disk space is not consumed when
    instances are started on a compute node, yet it is possible for
    scheduler to refuse to schedule an instance when combined disk usage of
    instances already running on the node exceeds total disk capacity
    reported by the hypervisor driver.

    Change-Id: I9718c727db205b6f2191f8435583391584e96e6e
    Closes-bug: #1332660
    Signed-off-by: Dmitry Borodaenko <email address hidden>

[Alexander Gubanov (ogubanov)]

I verified it on mos 5.1.1 - fixed