vmem drivers update stats gives invalid free/total size on query fail

[Ryan Lucio]

I discovered on testing CI today that occasionally if the backend array is under heavy load the per-minute stats query for free and total size may fail. This would cause the scheduler to invalidly stop allowing luns to be provisioned on the array.

I think that since this type of query failure is temporary, it should use the previously cached stats values instead.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/165763

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/165763
Committed: https://git.openstack.org/cgit/openstack/cinder/commit/?id=479b4c0dd3496f739e05082df0f2c675b34b7d0c
Submitter: Jenkins
Branch: master

commit 479b4c0dd3496f739e05082df0f2c675b34b7d0c
Author: Ryan Lucio <email address hidden>
Date: Thu Mar 19 03:17:17 2015 -0700

    Use cached values for stats on query failures for vmem drivers

    This change allows vmem drivers to return previously cached
    values for free_capacity_gb and total_capacity_gb if the queries
    to the backend for those stats succeed but do not contain
    the data. This can happen when the mgmt plane is under heavy load.

    Change-Id: I218a54897361d739c04963b4c36d93620be01fb3
    Closes-Bug: 1433990