Metrics

canette shows pod health and live CPU/memory usage for every app out of the box — no configuration required. Historical CPU/memory-over-time charts are opt-in and need a Prometheus-compatible query API, either a small bundled instance or your own.

What works with zero configurationCopy Anchor Link

• Pod health (ready/restart count) — read from the core Kubernetes Pods API, always available.
• Instant CPU/memory usage — read from metrics.k8s.io (metrics-server) when it's installed on the cluster. Degrades gracefully (no error, just an empty state) if it isn't.

Both appear as a stat-tile row on the app's Overview tab, and again on the dedicated Metrics tab.

Historical charts (metrics.enabled=true)Copy Anchor Link

CPU/memory-over-time charts need a Prometheus-compatible query API behind the scenes. Set metrics.enabled=true plus one of the two sources below — leave both unset and the UI silently falls back to the instant-only stat tiles, exactly as if metrics.enabled were false.

Bundled PrometheusCopy Anchor Link

A plain Deployment + ConfigMap (no prometheus-operator, no CRDs) that scrapes kubelet cAdvisor cluster-wide for container CPU/memory, retained for a short window (3 days by default) — a convenience chart, not a monitoring system:

metrics:
  enabled: true
  prometheus:
    bundled:
      enabled: true
      retention: 3d
      persistence:
        enabled: true
        size: 2Gi

The bundled instance scrapes via the Kubernetes apiserver-proxy path (/api/v1/nodes/<node>/proxy/metrics/cadvisor) rather than talking to each node's kubelet directly — this keeps its NetworkPolicy egress rule identical in shape to logstreamer's existing rule (Kubernetes API server only) and avoids kubelet TLS/anonymous-auth setup. Because of this, the ServiceAccount backing it (prometheus) needs a cluster-wide ClusterRole (nodes/proxy, nodes/metrics) — the only cluster-wide RBAC grant in the whole chart, everything else is namespace-scoped. Queries are always filtered to a single app's namespace and pod-name prefix at read time, so no app's data is exposed to another.

Set persistence.enabled: false to use an emptyDir instead of a PVC — cheaper, but history is lost on every pod restart or node drain.

External Prometheus-compatible endpointCopy Anchor Link

If you already run Prometheus, Thanos, Mimir, or VictoriaMetrics — they all implement the same PromQL HTTP query API — point canette at it instead of deploying a second instance:

metrics:
  enabled: true
  prometheus:
    externalUrl: https://prometheus.example.com

externalUrl takes precedence over prometheus.bundled whenever both are set. If your endpoint requires a bearer token (e.g. Grafana Cloud, multi-tenant Mimir), supply it via a Secret rather than a plain value:

kubectl create secret generic canette-prometheus-token \
  --namespace canette-system \
  --from-literal=token=...

metrics:
  prometheus:
    externalAuth:
      tokenSecretRef:
        name: canette-prometheus-token
        key: token

Where the charts show upCopy Anchor Link

• Overview tab — compact sparklines ride inline next to the existing CPU/memory stat tiles when historical data is available.
• Metrics tab — full-size charts with a hover crosshair, plus the same per-pod health list, grouped into sections so future metric families (e.g. a Traefik traffic adapter) can be added as a sibling section later without restructuring the page.

What's not implemented yetCopy Anchor Link

Request rate / status code / latency charts need a gateway-specific traffic adapter (Traefik first) — see canette/canette#168 for the staged rollout. metrics.gateway.provider exists in values.yaml as a placeholder but has no effect yet.