Incorrect labels after relating to grafana-agent rev 38

[przemeklal]

After relating hw-obs (tried rev 25 and rev 29) to grafana-agent rev 38, the labels are incorrect.

All hw-obs metric series use identical labels, especially problematic is that all of them use instance="localhost:10200" and juju_unit label doesn't exist anymore.

This makes it impossible to distinguish between metrics coming from different machines.

This is what the relevant metrics section in /etc/grafana-agent.yaml looks like:

metrics:
  configs:
  - name: agent_scraper
    remote_write:
    - tls_config:
        insecure_skip_verify: true
      url: http://<cos_traefik_ip>/cos-prometheus-0/api/v1/write
    scrape_configs:
    - job_name: hardware-observer_0_default
      metrics_path: /metrics
      static_configs:
      - labels:
          juju_application: hardware-observer
          juju_model: openstack
          juju_model_uuid: a2aaf764-adb9-43ae-9330-e5909e6891c1
        targets:
        - localhost:10200
  global:
    scrape_timeout: 60s
  wal_directory: /tmp/agent/data

This is a critical issue.

[przemeklal]

It works as before with revision 29 of g-agent. I suspect that this commit in g-agent canonical/grafana-agent-operator@3da25ab changed the topology and hw-obs will need to adapt to it.

[gabrielcocenza]

I'm closing this issue on this project because it will be fixed on canonical/grafana-agent-operator/issues/61