Node Monitor

This directory provides a local Prometheus stack for node and Docker container monitoring.

Start

chmod +x deploy-prometheus.sh
./deploy-prometheus.sh

If your server cannot pull from gcr.io, use the default mirror in .env.example:

cp .env.example .env
./deploy-prometheus.sh

If cAdvisor still cannot be pulled because of network timeout, start the core panel first:

./deploy-prometheus.sh --without-cadvisor

Access

Prometheus: http://YOUR_SERVER_IP:9090
Grafana: http://YOUR_SERVER_IP:3001
Grafana account: admin
Grafana password: admin123456

If you use a domain name, set GRAFANA_ROOT_URL to that public URL before starting Docker.

Grafana automatically provisions two dashboards in the Ping folder:

Ping 节点与 RabbitMQ 监控, used as the default home dashboard.
RabbitMQ 队列消费历史, a dedicated queue-history page that defaults to ping_logs_python.

Add Remote Nodes

Install node-exporter on the target server, allow the Prometheus server to access port 9100, then add another target group under the existing job_name: node in prometheus.yml:

- targets:
    - 192.168.1.10:9100
  labels:
    instance: app-server-01
    node_name: Hangzhou Telecom 01
    region: Zhejiang
    operator: Telecom

Keep remote targets in job_name: node; the bundled availability alert uses up{job="node"}.

Reload Prometheus after editing:

curl -X POST http://localhost:9090/-/reload

RabbitMQ Metrics

The Prometheus configuration automatically scrapes the Ping RabbitMQ metrics endpoint through host.docker.internal:15692. Deploy the files from the sibling rabbitmq directory so the rabbitmq_prometheus plugin and host port are enabled.

Verify both sides after deployment:

curl http://127.0.0.1:15692/metrics
curl http://127.0.0.1:9090/api/v1/query?query=up%7Bjob%3D%22rabbitmq%22%7D

The second response should contain a value of 1. Grafana can then query RabbitMQ metrics such as rabbitmq_queue_consumers and rabbitmq_queue_messages.

Notes

This compose file is intended for Linux servers. Docker Desktop on macOS can run the stack, but host-level node metrics are not equivalent to a real Linux host.

Make sure the server firewall or cloud security group allows inbound 9090 and 3001.

2.3 KiB Raw Blame History