Migrate metric queries from libvirt to OpenStack

[WizardBit]

Example, the compute overview dashboard has the following for scheduled VMs.

sum by(project_name) (libvirt_domain_info_meta{juju_application=~"$juju_application",juju_model=~"$juju_model",juju_model_uuid=~"$juju_model_uuid",juju_unit=~"$juju_unit",project_name=~"$project"})

Source

But, the cloud usage dashboard has the following for scheduled VMs.

sum(openstack_nova_running_vms{hostname=~"$hypervisor.*",juju_application=~"$juju_application",juju_model=~"$juju_model",juju_model_uuid=~"$juju_model_uuid",juju_unit=~"$juju_unit"})

Source

The first does not work without the libvirt exporter, and the second does. Making these changes where possible is beneficial due to the prometheus libvirt exporter being deprecated.

[syncronize-issues-to-jira]

Thank you for reporting your feedback to us!

The internal ticket has been created: https://warthogs.atlassian.net/browse/SOLENG-1485.

This message was autogenerated

[jneo8]

Thanks for bringing this up — the reasoning makes sense. Here are a couple of my thoughts:

• To enable the replacement, we’ll need to review all the dashboards to fully understand the scope.

• Regarding the metrics: while the values may appear the same in your example, their meanings differ slightly. One is collected via the OpenStack API, while the other comes from the libvirt exporter (which gathers metrics directly from the hypervisor). In some edge cases — for example, if the OpenStack API fails — the results might not match.

[Deezzir]

Looking at the dashboard, it seems like that the libvirt metrics are only used in compute.json dashboard. In my opinion, it seems more reasonable to migrate the metrics to OpenStack rather than trying to fix #179