Prometheus Node Exporter — Hardware and OS Metrics for Unix Systems

Introduction

Node Exporter is the standard way to feed host-level metrics into Prometheus. It reads data from Linux pseudo-filesystems (/proc, /sys) and exposes hundreds of hardware and kernel metrics on a single HTTP endpoint that Prometheus scrapes.

What Node Exporter Does

• Exposes CPU usage per core and per mode (user, system, idle, iowait)
• Reports memory, swap, and hugepage statistics from /proc/meminfo
• Tracks disk I/O, filesystem space, and mount-point health
• Monitors network interface bytes, packets, errors, and drops
• Provides thermal, power supply, and hardware sensor readings via sysfs

Architecture Overview

Node Exporter is a single Go binary that registers a set of collectors. Each collector reads from a specific Linux subsystem (procfs, sysfs, ethtool, etc.) and emits Prometheus-format metrics. Collectors are enabled or disabled via CLI flags. The binary listens on port 9100 and responds to GET /metrics with a text exposition of all enabled collectors.

Self-Hosting & Configuration

• Run as a systemd service with --web.listen-address=:9100
• Disable unneeded collectors with --no-collector.<name> flags
• Enable optional collectors like systemd, processes, or tcpstat explicitly
• Use --collector.textfile.directory to expose custom metrics from cron jobs
• Deploy as a Kubernetes DaemonSet using the Prometheus community Helm chart

Key Features

• 40+ built-in collectors covering CPU, memory, disk, network, and more
• Textfile collector: drop .prom files to expose custom batch metrics
• TLS and basic auth support for secure metric endpoints
• Low overhead: typically under 15 MB RSS and negligible CPU
• Pairs with Grafana's Node Exporter Full dashboard for instant visibility

Comparison with Similar Tools

• Telegraf — push-based, multi-output; Node Exporter is Prometheus-native pull
• cAdvisor — container-level metrics; Node Exporter covers the host OS
• collectd — plugin-based collector; Node Exporter is simpler, single-purpose
• Netdata — real-time dashboard included; Node Exporter is headless, pairs with Grafana

FAQ

Q: Does Node Exporter work on macOS or Windows? A: macOS is partially supported. Windows has a separate project: windows_exporter.

Q: How do I add custom metrics? A: Write a script that outputs Prometheus text format to a .prom file in the textfile directory. Node Exporter serves them automatically.

Q: What is the scrape performance impact? A: A typical scrape takes 50-200 ms and produces 1-3 MB of text. Impact on the host is negligible.

Q: Can I filter which metrics Prometheus collects? A: Yes. Use metric_relabel_configs in Prometheus or --collector.disable-defaults with explicit --collector.<name> flags.

Sources

• https://github.com/prometheus/node_exporter
• https://prometheus.io/docs/guides/node-exporter/