Set up observability & get the dashboards

The dashboards

Five dashboards, provisioned as code and brand-coloured:

• Resource Usage (CPU / GPU / memory)
• Signals Health (collection freshness)
• Analysis Overview (runs, findings)
• Insight Enrichment
• Signals Multi-Instance

Kubernetes: auto-imported via the Helm chart

The elevarq-unified chart ships the dashboards as a ConfigMap labelled grafana_dashboard: "1", so the standard Grafana dashboard sidecar (kube-prometheus-stack and friends) auto-discovers and imports them into an Elevarq folder — no manual import, no repository access. It is on by default:

grafanaDashboards:
  enabled: true            # default
  sidecarLabel: grafana_dashboard
  folder: Elevarq

This only publishes the dashboards; it does not deploy Grafana or Prometheus. Which panels show data depends on what your Prometheus scrapes (see below).

Docker Compose / your own Grafana: download

Grab the whole bundle, or individual files. Import a dashboard into Grafana with Dashboards → New → Import → Upload JSON.

Download the full observability bundle (.zip) — the five dashboards plus the example docker-compose.yml, prometheus.yml, and alert rules.

Or the individual files:

• Resource Usage (CPU / GPU / memory)
• Signals Health (collection freshness)
• Analysis Overview (runs, findings)
• Insight Enrichment
• Signals Multi-Instance
• Example docker-compose.yml — a local Prometheus + Grafana stack
• Example prometheus.yml — scrape config for Elevarq + Signals + node/GPU exporters
• analysis-alerts.yaml and signals-alerts.yaml — Prometheus alert rules

Point Prometheus at your metrics

The example prometheus.yml scrapes the Elevarq metrics endpoint (:9187), the Signals health endpoint, and node-exporter / DCGM for host and GPU resources. Two label dimensions the Signals dashboards drill down on:

• target — emitted by Signals; distinguishes the databases one collector monitors.
• signals_instance — distinguishes multiple Signals collector processes. Signals does not emit it; you add it per scrape target (the example sets signals_instance: primary). The multi-instance dashboard reads "no data" until your scrape config sets it.