In Django application, '/metrics' unaccessible when using Unix Domain Sockets

Im using the django_prometheus library which is a wrapper for prometheus_client library for my Django app. I am not running Prometheus, (or the entire applcaition in a Docker container), which might be part of my problem, though I dont see why this would be a requirement. The Django application is bound to Unix Domain Sockets,. NGINX is listening on default port 80 and re-routes traffic with the proxy_pass header.

NGINX

server {
    server_name hotname www.hostname ip;

[...]
    location /metrics{

        proxy_pass http://<unix_domain_socket>;
    }

    location / {
        proxy_pass http://unix:<unix_domain_socket>;
        
}

prometheus

<default prom config>
[...]
    static_configs:
      - targets: ["localhost:9090"]

  - job_name: "django-app"
    scrape_interval: 10s
    static_configs:
      - targets: ["localhost] # nothing works (IP, etc..) and prometheus attempts to use port 80 with a predefined scheme of http://

  - job_name: "node"
    scrape_interval: 10s
    static_configs:
      - targets: ["localhost:9100"]

Solution

By removing the unix domain socket in NGINX and going back to TCP/IP that gunicorn uses by default, Prometheus was able to successfully pickup metrics from /metrics

Is there something Im missing? I feel like the only other options are changing the whole infrastructure ( to docker), or try and use a log based-metrics.

先从 issue 中所示的 NGINX location 块和 Prometheus scrape target 配置开始，然后将 Unix domain socket 设置与正常工作的 Gunicorn TCP 设置进行比较。确定 metrics endpoint 是否支持此部署路径，并记录经过验证的配置或所需的限制。