VMware VMDK driver should use host mount count as the main criterion for datastore selection

[Vipin Balachandran]

Currently the VMDK driver picks a datastore for volume creation based on space utilization. The shell VM corresponding to the volume is migrated to a different host and datastore if the volume is attached to an instance created in a host which cannot access the volume's current datastore. These migrations could be minimized if a datastore connected to multiple hosts is picked at the time of volume creation.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/56350

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/56350
Committed: https://git.openstack.org/cgit/openstack/cinder/commit/?id=b85cc49588c9f40e091801af33da8c7575058ec0
Submitter: Jenkins
Branch: master

commit b85cc49588c9f40e091801af33da8c7575058ec0
Author: Vipin Balachandran <email address hidden>
Date: Wed Nov 13 20:33:39 2013 +0530

    VMDK:Using host mount info for datastore selection

    Currently the VMDK driver picks a datastore for volume creation based
    on space utilization. The shadow VM corresponding to the volume is
    migrated to a different host and datastore if the volume is attached
    to an instance created in a host which cannot access the volume's
    current datastore. To reduce these migrations, this fix selects a
    datastore which is connected to maximum number of hosts, provided
    there is enough space to accommodate the volume. Ties are broken
    based on space utilization; datastore with low space utilization
    is preferred.

    Closes-Bug: #1250816
    Change-Id: I5ecf22d782857fca0889799229465c49afe188a0