monitoring

Metrics stack for ClashPerk.

Folder	Runs on	Public URL
victoria-metrics/	Docker Compose on a small VM (the "metrics host"), port 8428	https://metrics.clashperk.com (basic auth)
grafana/	a container platform, built from grafana/Dockerfile	https://grafana.clashperk.com (Google login, clashperk.com only)

VictoriaMetrics stores every sample; Grafana stores only users and settings and queries VictoriaMetrics through its public URL.

Deploy

• VictoriaMetrics: the host holds ~/victoria-metrics/ plus a .env (mode 600) made from .env.example. With $HOST as its SSH alias:

  scp victoria-metrics/docker-compose.yml $HOST:victoria-metrics/
  ssh $HOST 'cd ~/victoria-metrics && docker compose up -d'

• Grafana: rebuild and redeploy the image; see grafana/README.md.

Secrets never go in this repo.

Who pushes metrics

Services push to https://metrics.clashperk.com/api/v1/import/prometheus with extra_label=job=<service>; VictoriaMetrics can't reach the other servers to scrape them. Currently: the API and the worker (clashperk-backend MetricsPushModule, job=api / job=worker, worker forks add fork) the Discord bot (clashperk-bot, job=bot, one series per cluster), the image service (image-service, job=images), and the Clash API proxy (coc-api-proxy, job=proxy; it sits on the same private network, so it pushes to the private address). Keep label values bounded: route patterns, never raw URLs or tags.