Home Original page

GitHub - Enscend/BasicDockerMonitor: Basic Docker monitoring setup using public Docker Compose and Docker images from Grafana, Prometheus, and CAdvisor

Initial Setup

Note: ChatGPT did most of the work on this one...

  1. Start it up using Docker Compose via the _runMonitor.bat or _runMonitor.sh scripts (depending on your system type), because we need to set up as environment variable for the system root.
  • assumes you already have Docker properly set up on your system
  • docker-commpose.yml is set up to restart:always so it will automatically come back up when you restart the computer

Windows

C:\your\repo\folder\> _runMonitor

Linux

user@system:~/your/repo/folder$ sudo chmod +x ./_runMonitor.sh
user@system:~/your/repo/folder$ ./_runMonitor.sh
  1. Navigate to: https://localhost:3000 to load the Grafana UI
  2. Change the default password for Grafana
  3. From the Data sources link in the left-side nav panel, click Add new data source to add Prometheus as a data source
  • Prometheus server URL *: http://prometheus:9090
  • Prometheus type: Prometheus
  • Prometheus version: > 2.50.x
  • Click Save & Test
  1. From the Dashboards link in the left-side nav panel, click New then Import
  • Enter '893' (one that ChatGPT told me about) for Docker monitoring, then click Load
  • Select the Prometheus data source that you created above, then click Import
  1. From the Dashboards link in the left-side nav panel, select the Docker and system monitoring dashboard to view your stats

vLLM Metrics

This stack is pre-configured to scrape metrics from a vLLM instance running on the host at port 8000.

Prerequisites:

  • Start vLLM with metrics enabled (the flag varies by version): 
    --enable-metrics          # older versions
--prometheus-port 8000    # some versions expose on a separate port
    Verify the endpoint is live: curl http://localhost:8000/metrics

After starting the stack:

  1. Open http://localhost:9090/targets — the vllm job should show status UP
  2. In Prometheus, query a metric like vllm:num_requests_running to confirm data is flowing
  3. In Grafana, import a vLLM dashboard: Dashboards > New > Import, then search the Grafana dashboard library for "vLLM"

Linux note: The extra_hosts: host.docker.internal:host-gateway entry in docker-compose.yml is what allows Prometheus (inside the monitoring bridge network) to reach the vLLM process on the host network. On Docker Desktop (Windows/Mac) this resolves automatically.