Dashboards Are Outputs, Scrape Configs Are the Work
Most monitoring guides start with a Grafana screenshot and skip straight to dashboard import IDs. This one starts with the Prometheus scrape model, because that is where the data comes from and where things actually go wrong. Prometheus is a pull-based time-series database: it polls configured targets on a schedule, collects numeric metrics exposed at a /metrics HTTP endpoint, and stores them with timestamps. Grafana queries Prometheus and renders the results. The work is in knowing what to instrument and writing scrape configs that collect the right data.
15Second Scrape Interval
Set Up Your Exporters
Exporters are small programs that translate system or application state into Prometheus-format metrics at /metrics. Install three for a complete home lab picture. node_exporter exposes CPU, memory, disk, and network metrics for the host machine — run it a
Write a docker-compose.yml for the Full Stack
Create a monitoring directory: mkdir -p ~/docker/monitoring. The compose file defines four services: prometheus, grafana, node_exporter, and cadvisor. Prometheus mounts a named volume for its time-series data directory and a bind-mounted prometheus.yml co
Configure Prometheus Scrape Jobs
In prometheus.yml, the scrape_configs section lists your targets. Each job has a name and a list of static targets — the host:port of each exporter. Add jobs for node (targeting node_exporter:9100), cadvisor (targeting cadvisor:8080), and blackbox (target
Write Basic PromQL Queries
PromQL is Prometheus's query language. The simplest query is a metric name: up returns 1 for each scrape target currently responding and 0 for each one that is not. To see CPU usage, use rate(node_cpu_seconds_total{mode!="idle"}[5m]) to calculate the per-
Import a Grafana Dashboard and Set Up Alerting
In Grafana, go to Dashboards and choose Import. Enter dashboard ID 1860 to import the popular Node Exporter Full dashboard, which provides a comprehensive view of your Pi's resources without writing any queries. Point the dashboard's data source at your P
What Comes Next
Metrics tell you what your applications are doing right now — CPU is high, memory is growing, a container went down at 3:17 AM. But when something goes wrong and you need to understand what happened leading up to it, you need logs. Centralizing logs from every container and system service into one queryable store is what makes post-incident investigation possible.
