Wrong samples of the "cpu_util" meter

[Alexander Rubtsov]

=== Environment ===
MOS: 9.2 (ISO build #562)
Ceilometer: 1:6.1.4-5~u14.04+mos27
Hypervisor: Libvirt+QEMU

=== Description ===
There are samples of "cpu_util" with negative value

=== Steps to reproduce ===
1. Deploy an environment with Ceilometer and Libvirt+QEMU/KVM hypervisor backend
2. Launch a new VM instance
3. Keep it running for a while so that Ceilometer will collect a couple of samples
4. Reboot the VM instance
5. Wait for a while so that Ceilometer will collect the next sample

=== Actual result ===
ceilometer sample-list -m cpu_util -q 'resource_id=<VM_uuid>'
There is a negative value of the sample:
http://paste.openstack.org/show/EGeib43DZMqmdlsEvtsT/

=== Expected result ===
ceilometer sample-list -m cpu_util -q 'resource_id=<VM_uuid>'
A real value has been collected:
http://paste.openstack.org/show/jjB4oTwmK3nkI0RXUSSo/

=== Notes ===
Also, the extra samples are observed sometimes
Their timestamps differ from the main ones on several ms
For example:
http://paste.openstack.org/show/H75MAioBJvCyN3fCqnJj/
Probably, it's related to wrong logic of CPU util calculation as well

[Alexander Rubtsov (arubtsov)]

sla1 for 9.0-updates

[Denis Meltsaykin (dmeltsaykin)]

After a conversation with Ildar Svetlov it turned out that this is a known issue and there was a try to fix it https://review.fuel-infra.org/#/c/35876, unfortunately it has not been finished. We'll check what we can do.

[Fuel Devops McRobotson (fuel-devops-robot)]

Reviewed: https://review.fuel-infra.org/35876
Submitter: Pkgs Jenkins <email address hidden>
Branch: 9.0/mitaka

Commit: 297a971636f2241d8d9b920ab179519fc1230d6b
Author: Ildar Svetlov <email address hidden>
Date: Wed Sep 6 12:33:55 2017

Skip negative cpu_util metrics

Closes-Bug: #1713714

Change-Id: I5807e421933b776a019b0f21a1a1124f385b32fa

[Dmitry (dtsapikov)]

Verified on 9.2+mu3