Add 'aggregate' label to nova metrics

[openstack-exporter]

Add 'aggregate' label to nova metrics

[niedbalski]

@prolane

In my opinion a single label named 'aggregates', which can be a space/comma separated list of aggregates on which a specific hypervisor belongs to should suffice to solve this issue as you can filter by a specific label and sum the vcpus_available. Any drawbacks on using this approach?

[prolane]

Thats certainly a possibility @niedbalski. I guess if we make the space/comma separated list in alphabetical order, we make sure we don't end up with lots of different combinations. For example, you don't get one hypervisor which has the 'aggregates' label value set to "az-west,ssd-hosts,shared-cluster" and one other set to "ssd-hosts,az-west,shared-cluster".

Just one drawback of this approach is, if you have hypervisors being part of more than one aggregate, you can not group by one aggregate. For example, with the example aggregates label values above, I won't be able to group by just the ssd-hosts. Not without making hard coded dashboards in Grafana for example.

Ideally, you would want to do something like this:
sum(vcpus_available) by(aggregates)

And make rules (for alerting), like this:
sum(vcpus_available) by(aggregates) < 20

This will still work for the collection of hypervisors which all have the same combination of aggregates (e.g. "az-west,ssd-hosts,shared-cluster"), but if I then would want to alert on when vcpus are running low on just the ssd-hosts, I would have to make it hardcoded like:
sum(vcpus_available{aggregates=~".*ssd-hosts.*"}) by(aggregates) < 20

Perhaps this is fine, as I guess usually teams don't often add new aggregates?

[prolane]

This PR does the job: #75

Happy X-mas holidays! Cheers!

[niedbalski]

Merged #75 , Thank you very much for your amazing contributions @prolane !