Instance metrics are not enabled

[Lucian Petrut]

Hyper-V's metric collection API is extremely inconsistent.

As opposed to disk metrics, network metrics have to be enabled whenever the vm starts. Attempting to do so while the vm is shut off will fail. This option also gets reset when the vm is rebooted.

CPU/memory metrics need to be re-enabled after instance reboots as well. This looks like a Hyper-V bug (which affects 2012R2 as well as WS 2016), since soft shutdowns initiated from the host side don't seem to affect the metrics.

At the moment, networking-hyperv enables port metrics but only when the ports are plugged. For this reason, we won't be able to collect metrics after instance reboots.

Also, we're not enabling metrics for OVS ports at all. Metrics for Cinder volumes aren't enabled either. Note that enabling metrics for passthrough disks isn't supported.

We'll have to workaround those Hyper-V limitations at the Nova driver level.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/605359
Committed: https://git.openstack.org/cgit/openstack/compute-hyperv/commit/?id=8c44c7a960f99b582c733118b7c2d24dcca51b79
Submitter: Zuul
Branch: master

commit 8c44c7a960f99b582c733118b7c2d24dcca51b79
Author: Lucian Petrut <email address hidden>
Date: Wed Sep 26 11:27:17 2018 +0300

    Ensure instance metrics get collected

    The Hyper-V metric collection API has the following limitations:
    * some metrics get disabled when the instance is rebooted (network,
    cpu and memory)
    * network metrics cannot be enabled while the vm is stopped
    * metrics cannot be collected for passthrough disks
    * some times, the metric counters are reset when the vms are rebooted.
    We can't do much about this Hyper-V bug, which is unfortunately
    affecting the reliability of the collected data.

    At the same time, OVS ports and Cinder volumes aren't handled at all.

    To overcome those issues, this patch brings the following changes:
    * re-enable vm metrics each time the vms boot
    * enable vm metrics after live migrations
    * enable metrics metrics when binding OVS ports (networking-hyperv
      will handle 'hyperv' ports)
    * enable disk metrics when attaching SMB Cinder volumes, logging a
      warning for passthrough disks

    Closes-Bug: #1794456
    Depends-On: Ibe7409b392b91204f0cea2bdd32c86ec03d06b67
    Depends-On: Iaba32ca94b2c09661f150fa4b12b2c7eb83ff6d9

    Change-Id: I94f2be20b02531d3ceabdef4272207e37ef34930

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/compute-hyperv 9.0.0.0rc1 release candidate.