Prometheus Node Exporter

The default Prometheus exporter for host metrics. Run it on every machine you want to monitor, point Prometheus at it, get CPU, memory, disk, network, filesystem, and kernel stats for free.

GitHub · Docs

What it exposes

The full list is long. The categories that matter for most services (USE method):

CPU: node_cpu_seconds_total, node_load1/5/15
Memory: node_memory_MemTotal_bytes, node_memory_MemAvailable_bytes
Disk: node_disk_io_now, node_filesystem_avail_bytes
Network: node_network_receive_bytes_total, node_network_transmit_bytes_total
Kernel: node_vmstat_*, node_uname_info
Filesystem: node_filesystem_files, node_filefd_*

How to run it

Three common shapes:

Bare metal / VM: systemd service, started on boot, scrapes localhost:9100/metrics
Docker compose: one container per host, network mode host so it can see the host's filesystem and network
Kubernetes: DaemonSet so one pod runs on every node

Compose example (used in the linko repo alongside the app's /metrics):

node-exporter:
  image: prom/node-exporter:latest
  network_mode: host
  pid: host
  volumes:
    - /proc:/host/proc:ro
    - /sys:/host/sys:ro
    - /:/rootfs:ro
  command:
    - '--path.procfs=/host/proc'
    - '--path.sysfs=/host/sys'
    - '--path.rootfs=/rootfs'
    - '--collector.filesystem.mount-points-exclude=^/(sys|proc|dev|host|etc)($$|/)'

network_mode: host + the /proc and /sys mounts are the important bits. Without them the exporter only sees its own container, which is useless for host metrics.

Scrape config

Tell Prometheus to scrape it:

scrape_configs:
  - job_name: 'node'
    static_configs:
      - targets: ['localhost:9100']

In the Boot.dev metrics note, the same compose file runs Prometheus, the app, and node-exporter together.

What to alert on

The standard alerts from the USE method:

Signal	Alert
`up == 0`	Host is down
`predict_linear(node_filesystem_avail[6h], 24*3600) < 0`	Disk will fill in 24h
`node_load5 > N` (N = cores * 1.5)	CPU is over-saturated
`node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes < 0.1`	Memory pressure

Same pattern as Boot.dev alerts, but at the host layer.

When to use something else

Node Exporter is for Linux hosts. For other targets:

Windows: windows_exporter
Containers / k8s pods: cAdvisor, built into kubelet
NVIDIA GPUs: dcgm-exporter
Network gear: snmp_exporter
Anything else: most things have a community exporter on the Prometheus exporters page