// the find
prometheus/node_exporter
Exporter for machine metrics
The de facto standard Prometheus exporter for Linux/Unix host metrics — CPU, memory, disk, network, filesystem, and about 50 other collectors pulled from /proc and /sys. It's the first thing you install when setting up Prometheus on any machine, and the collector breadth is genuinely impressive. If you're running Prometheus, you're running this.
The pluggable collector architecture is well-executed — each collector is independently toggleable at startup, and the include/exclude filter flags let you trim cardinality without rebuilding anything. The textfile collector is a smart escape hatch: anything node_exporter doesn't cover natively can be injected via a .prom file, which is how most custom host-level metrics end up in the pipeline. Cross-platform support is real and not faked — separate per-OS files with build tags, fixture-based tests for each platform, and CI that actually runs BSD checks. The sysctl collector is a recent addition that's more useful than it looks: you can expose arbitrary kernel tuning values as metrics without writing any code.
Container deployment is genuinely awkward — you need host networking, host PID namespace, and a bind mount of / just to monitor what you actually want. The README documents this but it's a footgun for anyone dropping it into a Kubernetes DaemonSet without reading carefully first. TLS support is still marked experimental after years, which means anyone who wants HTTPS between Prometheus and node_exporter is either skipping it or trusting an unstable config path. The systemd collector is disabled by default because of performance concerns, but system health monitoring without it is a real gap — you're flying blind on service state unless you enable something the project itself warns you about. Several collectors (tcpstat, perf) have documented performance issues in high-load scenarios with no clear fix beyond 'don't use them on busy machines.'