inference-harness/nginx/nginx.conf

events {
    worker_connections 1024;
}

http {
    include /etc/nginx/mime.types;
    default_type application/octet-stream;
    sendfile on;
    keepalive_timeout 65;

    resolver 127.0.0.11 valid=30s;

    map $host $dashboard_ui_url {
        default http://harness-dashboard:3000;
    }
    map $host $litellm_backend_url {
        default http://harness-litellm:4000;
    }

    # ════════════════════════════════════════════════════════════
    # Server :80 — Lean single-layer entrypoint
    #   All API paths route directly to LiteLLM.
    #   harness-router fully deprecated.
    # ════════════════════════════════════════════════════════════
    server {
        listen 80;

        # Authentik OIDC subrequest
        location /authentik/auth {
            internal;
            proxy_pass https://auth.sysloggh.net/outpost.goauthentik.io/auth/nginx;
            proxy_pass_request_body off;
            proxy_set_header Content-Length "";
            proxy_set_header X-Original-URL $scheme://$http_host$request_uri;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        }

        # ── LiteLLM API (replaces router /v1/) ──
        location /v1/ {
            proxy_pass $litellm_backend_url;
            proxy_http_version 1.1;
            proxy_set_header Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
            proxy_set_header Authorization $http_authorization;
            proxy_connect_timeout 10s;
            proxy_read_timeout 600s;
            proxy_buffering off;
        }

        # ── LiteLLM admin (replaces router /admin/) ──
        location /admin/ {
            proxy_pass $litellm_backend_url;
            proxy_http_version 1.1;
            proxy_set_header Host $host;
            proxy_set_header Authorization $http_authorization;
            proxy_read_timeout 600s;
        }

        # ── LiteLLM stream ──
        location /stream {
            proxy_pass $litellm_backend_url/stream;
            proxy_http_version 1.1;
            proxy_set_header Host $host;
            proxy_set_header Upgrade $http_upgrade;
            proxy_set_header Connection "upgrade";
            proxy_buffering off;
        }

        # ── API passthrough ──
        location /api/ {
            proxy_pass $litellm_backend_url/;
            proxy_http_version 1.1;
            proxy_set_header Host $host;
        }

        # ── Dashboard ──
        location /dashboard/ {
            proxy_pass $dashboard_ui_url/;
            proxy_http_version 1.1;
            proxy_set_header Host $host;
        }

        # ── GPU Fleet Dashboard ──
        location /gpu/ {
            proxy_pass http://192.168.68.24:9100/;
            proxy_http_version 1.1;
            proxy_set_header Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_read_timeout 60s;
        }

        # ── LiteLLM redirect ──
        location = /litellm {
            return 301 /litellm/;
        }

        # ── Health: redirect /health (auth-required) → /health/liveliness (no-auth) ──
        location = /litellm/health {
            return 301 /litellm/health/liveliness;
        }

        # ── LiteLLM static assets ──
        location /litellm-asset-prefix/ {
            proxy_pass $litellm_backend_url;
            proxy_http_version 1.1;
            proxy_set_header Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_connect_timeout 10s;
            proxy_read_timeout 60s;
        }

        # ── LiteLLM admin UI and API (strips /litellm prefix) ──
        location /litellm/ {
            rewrite ^/litellm(/.*)$ $1 break;
            proxy_pass $litellm_backend_url;
            proxy_http_version 1.1;
            proxy_set_header Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
            proxy_set_header Upgrade $http_upgrade;
            proxy_set_header Connection "upgrade";
            proxy_buffering off;
proxy_read_timeout 600s;
            proxy_set_header Authorization $http_authorization;
        }

        # ── Auth proxy to Authentik ──
        location /application/o/ {
            proxy_pass https://192.168.68.11;
            proxy_ssl_verify off;
            proxy_set_header Host auth.sysloggh.net;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_connect_timeout 10s;
            proxy_read_timeout 30s;
        }

        # ── Prometheus metrics (LiteLLM exposes at /metrics) ──
        location /metrics {
            proxy_pass $litellm_backend_url/metrics;
            proxy_http_version 1.1;
            proxy_set_header Host $host;
        }

        # ---- Legacy /health/unified alias (harness-router decommissioned) ----
        #      Retained intentionally: 5+ live GPU monitor contracts poll
        #      this path every 15s and follow the 301 to /gpu/gpu-data.
        location /health/unified {
            return 301 /gpu/gpu-data;
        }

        # ── Health: no-auth LiteLLM liveliness ──
        location /health {
            proxy_pass $litellm_backend_url/health/liveliness;
            proxy_http_version 1.1;
            proxy_set_header Host $host;
        }

                
        # ── UI / Docs convenience redirects (additive 2026-09-11) ──
        #   Canonical app path is /litellm/. These 301s make bare
        #   /ui/ and /docs resolve. No new auth surface; no OIDC impact.
        location = /ui { return 301 /litellm/ui/; }
        location /ui/ { return 301 /litellm$request_uri; }
        location = /docs { return 301 /litellm/docs; }
        location = /docs/ { return 301 /litellm/docs; }

        # ── 404 for everything else ──
        location / {
            return 404;
        }
    }
}