Update service.json for new agent metrics names

[mathmarchand]

Nova, Cinder and Neutron agent metric names changed to openstack_agent_total from openstack_agent.

[mathmarchand]

Related to this issue .

[gabrielcocenza]

Thanks for this patch @mathmarchand. May I ask you some things?
1 - Contribute to the upstream to fix the Readme
2 - Share a print in the PR with the prometheus metrics or the dashboard properly
3 - Update the Prometheus alert rules in sunbeam. (We sync the alert rules from there)

[mathmarchand]

Hi @gabrielcocenza,

Some more context, it seems that addition of _total is not from upstream openstack-exporter code.:

The Prometheus Go client library enforces the OpenMetrics specification, which requires that all counter-type metrics must end in _total. So when the metric type is declared as CounterValue, the Go client automatically appends _total to the metric name at scrape time.

This means:

The metric is defined in code as agent_state
The exporter registers it as openstack_nova_agent_state
Prometheus scrapes it as openstack_nova_agent_state_total

Basically if prometheus is on version 2.x+, the _total metric need to be used, if not the old metric need to be used. Since, the COS Lite 1/stable is EOL and we should be using 2/stable until the new LTS version is released, this _metrics is needed.

I will update the query so that it could work in both situations.

[mathmarchand]

As requested, here's a output of the adjusted query in Grafana.

[Deezzir]

@mathmarchand The addition of _total suffix is coming from opentelemetry-collector. We pushed the change to the collector to disable the normalization for now. Please see canonical/opentelemetry-collector-k8s-operator#238 and canonical/opentelemetry-collector-k8s-operator#232. The change you're proposing shouldn't be needed anymore.

P.S. If we merge the PR, the backward compatibility with grafana-agent will be broken.