ceph driver deletes can cause excessive exceptions

[Walt Boring]

When issuing lots of delete calls in parallel for the ceph driver, it can issue lots of ImageNotFound exceptions. When cinder does a delete_volume it will ask the driver to get_volume_stats, which for the rbd driver can result in scans of every volume in the cluster.
There is a race condition where parallel deletes can cause an image in the pool to be removed after another delete call has gotten it in it's list for scans.

_get_usage_info() gets a list of all volumes to scan for their size here:
https://github.com/openstack/cinder/blob/stable/pike/cinder/volume/drivers/rbd.py#L384-L403

The exception is Logged here:
https://github.com/openstack/cinder/blob/stable/pike/cinder/volume/drivers/rbd.py#L144

In the case of getting volume stats, the exception is later ignored here:
https://github.com/openstack/cinder/blob/stable/pike/cinder/volume/drivers/rbd.py#L402-L403

We should add a new param to the RBDVolumeProxy to silence the logging of the ImageNotFound exception, since it's going to be debug logged and ignored for this case.

[Walt Boring (walter-boring)]

This was already addressed.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/pike
Review: https://review.opendev.org/680764

[Walt Boring (walter-boring)]

This bug is for backporting the existing fix to pike.

https://github.com/openstack/cinder/commit/71584d6852a6dec8d8d04c8f3cbddf0f114777af#diff-439c80f9f706c9c6c8fe90266cde5c40

https://review.opendev.org/#/c/680764/

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/680764
Committed: https://git.openstack.org/cgit/openstack/cinder/commit/?id=d9b4a9af821e77057d890151003014e68ddcf7ca
Submitter: Zuul
Branch: stable/pike

commit d9b4a9af821e77057d890151003014e68ddcf7ca
Author: Jon Bernard <email address hidden>
Date: Wed May 9 14:43:56 2018 -0400

    RBD: remove redundant exception log to reduce noise

    If RBDVolumeProxy fails to open the underlying image it reraises the
    exception, so there is no need to log it unconditionally. The caller
    can gracefully handle this case without alarming the admin.

    Closes-Bug: #1843074
    Change-Id: If5a9d014dbdd63a6e4659d700671c6ac6496b96a
    (cherry picked from commit 71584d6852a6dec8d8d04c8f3cbddf0f114777af)