Skip to content

USER GUIDE

Prometheus metrics

Purpose

Scrape hook execution, event throughput, routing-rule counts, and netlink health for fleet observability — the same HTTP server as the REST API on :9090.

When to use it

  • Alert when netevd_script_failures_total rises after a hook deploy
  • Dashboard event rates per interface/backend
  • Detect silent daemon failure (uptime gauge flatlines)

How to get there

  • Enable: metrics.enabled: true in /etc/netevd/netevd.yaml
  • Endpoint: http://<host>:9090/metrics on the API server (Prometheus text exposition)
  • metrics.port (default 9091) is stored in the config and logged at startup; the daemon serves /metrics on api.port

Operate from CLI

  1. Confirm metrics are enabled:
grep -A3 '^metrics:' /etc/netevd/netevd.yaml
curl -sf http://127.0.0.1:9090/metrics | head -20
  1. Inspect key series locally:
curl -sf http://127.0.0.1:9090/metrics | grep -E '^netevd_(info|events_total|script_|routing_rules|ebpf_)'
  1. Prometheus scrape config (replace <host> with your edge/management target):
scrape_configs:
  - job_name: netevd
    scrape_interval: 30s
    static_configs:
      - targets: ['<host>:9090']
        labels:
          role: network-events
  1. Example alert rules (hook failures):
groups:
  - name: netevd
    rules:
      - alert: NetevdHookFailures
        expr: increase(netevd_script_failures_total[5m]) > 0
        for: 2m
        labels:
          severity: warning
      - alert: NetevdDown
        expr: absent(netevd_info)
        for: 5m
        labels:
          severity: critical
  1. Remote check from your workstation:
curl -sf http://<host>:9090/metrics | grep netevd_uptime_seconds

  1. Empty / fail: Connection refused → metrics.enabled: false or firewall; series missing → daemon just started or metrics init failed (see journalctl -u netevd).

  2. Success: netevd_info{version="…"} present; counters increment when you bounce a link and routable hooks run.