feat(profile): wire COS observability into kubernetes and machine scaffolds

[tonyandrewmeyer]

What this does

Extends both the kubernetes and machine profiles so that charmcraft init
produces a charm pre-wired for Canonical Observability Stack (COS) integration.
Each profile is independent; six logical commits cover the work.

---

Kubernetes profile (--profile=kubernetes)

Charm-libs declared in charmcraft.yaml (run charmcraft fetch-libs after init):

• prometheus_k8s.prometheus_scrape v0
• loki_k8s.loki_push v1
• grafana_k8s.grafana_dashboard v0

Relations added to charmcraft.yaml:

provides:
  metrics-endpoint:
    interface: prometheus_scrape
  grafana-dashboard:
    interface: grafana_dashboard
requires:
  logging:
    interface: loki_push

src/charm.py — three providers instantiated in __init__:

self._metrics_endpoint = MetricsEndpointProvider(
    self, jobs=[{"static_configs": [{"targets": ["*:8080"]}]}]
)
self._log_forwarder = LogForwarder(self)
self._grafana_dashboard = GrafanaDashboardProvider(self)

src/grafana_dashboards/overview.json — starter dashboard with:

• up stat panel (prometheus scrape health) driven by $juju_application
• markdown panel with getting-started guidance

Unit tests (ops.testing.Context):

• test_metrics_endpoint_relation_joined — verifies scrape_jobs with port 8080 is written to app data
• test_grafana_dashboard_relation_joined — verifies dashboards key is populated in app data
• test_logging_relation_joined — verifies LogForwarder handles the requirer-side event without error

Integration tests: skipped stubs showing the jubilant + COS pattern for each relation.

---

Machine profile (--profile=machine)

Charm-lib declared in charmcraft.yaml:

• cos_agent.cos_agent v0

Relation added:

provides:
  cos-agent:
    interface: cos_agent

src/charm.py — single provider that bundles metrics, logs, and dashboards:

self._cos_agent = COSAgentProvider(
    self,
    metrics_endpoints=[{"path": "/metrics", "port": 8080}],
    dashboard_dirs=["./src/grafana_dashboards"],
)

src/grafana_dashboards/overview.json — identical structure to the kubernetes starter dashboard (grafana-agent picks this up via the cos-agent relation and ships it to Grafana).

Unit tests (ops.testing.Context):

• test_cos_agent_relation_joined — verifies config key with metrics_scrape_jobs is written to unit data
• test_cos_agent_relation_joined_dashboard_registered — verifies dashboards are included in config

Integration tests: skipped stubs covering the grafana-agent co-deployment pattern.

---

How to activate (post-init)

charmcraft init --profile=kubernetes  # or --profile=machine
charmcraft fetch-libs                  # downloads the charm-libs into lib/
# inspect PYDEPS in each fetched lib, add any extras to pyproject.toml dependencies
tox -e unit                            # unit tests pass once libs are fetched

To activate COS monitoring after deploying:

Kubernetes:

juju integrate <app>:metrics-endpoint prometheus-k8s:metrics-endpoint
juju integrate <app>:grafana-dashboard grafana-k8s:grafana-dashboard
juju integrate <app>:logging loki-k8s:logging

Machine:

juju deploy grafana-agent
juju integrate <app>:cos-agent grafana-agent:cos-agent
# then cross-model relations to COS lite

COS integration test limitation

Full COS integration tests (marked @pytest.mark.skip) require a live COS lite
stack and (for machine) the grafana-agent machine charm. These are documented
in-template with setup instructions. The skipped stubs show the expected jubilant
pattern; developers can enable them once their environment has COS deployed.

Out of scope

• Tracing (separate routine; quick to add via opentelemetry_sdk lib but not included here)
• Alert rules directories (prometheus_alert_rules/, loki_alert_rules/) — developers add these as needed
• 12-factor profiles and test-* overlays

---

Generated by Claude Code