abiba-bot/inference-harness
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

feat: 2-layer architecture foundation + agent migration

Browse Source 
Router v2:
- Atomic GPU slot booking (Redis Lua — closes TOCTOU race)
- Circuit breaker with failure threshold (3 failures/120s → 60s cooldown)
- GPU health scoring with configurable weights (VRAM 40% + temp 30% + load 30%)
- X-Usage-Tokens header for LiteLLM spend tracking
- /health/unified endpoint (aggregates all layers)
- Strict explicit model passthrough (no silent fallback)
- syslog-auto → auto routing fix

Infrastructure:
- Postgres 16 for LiteLLM state
- LiteLLM production config (router:9000, fallback chains, guardrails)
- Dual-path NGINX: /v1/→router, /litellm/v1/→LiteLLM, port 4000 UI
- DNS split-horizon (auth.sysloggh.net → 192.168.68.11)
- 6 LiteLLM virtual keys for agent cutover

Deployed to CT 116, all 6 containers healthy.
This commit is contained in:
Abiba
2026-06-17 22:40:33 +00:00
parent 776343f2ab
commit fd3c2a575a
4 changed files with 498 additions and 90 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nginx/nginx.conf
+193 -49
View File
@@ -1,18 +1,10 @@
worker_processes auto;
error_log /var/log/nginx/error.log warn;
pid /var/run/nginx.pid;
events { worker_connections 1024; }
events {
worker_connections 1024;
}
http {
include /etc/nginx/mime.types;
default_type application/octet-stream;
log_format main '$remote_addr - $remote_user [$time_local] "$request" '
'$status $body_bytes_sent "$http_referer" '
'"$http_user_agent" rt=$request_time';
access_log /var/log/nginx/access.log main;
error_log /var/log/nginx/error.log;
sendfile on;
keepalive_timeout 65;
@@ -20,18 +12,33 @@ http {
upstream dashboard_ui { server dashboard:3000; }
upstream litellm_backend { server litellm:4000; }
# Detect Cloudflare Tunnel requests (cloudflared always sets CF-Connecting-IP).
# Direct LAN browser access to :4000 has no such header -> redirect to canonical https,
# preventing the cross-origin localStorage footgun that traps the UI at the login page.
map $http_cf_connecting_ip $is_cloudflared {
default 1; # any non-empty value = request came through Cloudflare
"" 0; # empty = direct access
}
# ════════════════════════════════════════════════════════════════
# Server :80 — existing harness entrypoint
# dashboard (/), router API (/v1/, /admin/, /stream, /api/, /metrics),
# router fallback, LiteLLM UI via /litellm/ prefix, health
# ════════════════════════════════════════════════════════════════
server {
listen 80;
# Security headers
add_header X-Content-Type-Options nosniff always;
add_header X-Frame-Options SAMEORIGIN always;
add_header X-XSS-Protection "1; mode=block" always;
# 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;
}
# Disable buffering for SSE streams
proxy_buffering off;
# API through router
# 2-Layer: /v1/ → router (existing keys work unchanged)
location /v1/ {
proxy_pass http://router_api;
proxy_http_version 1.1;
@@ -41,6 +48,16 @@ http {
proxy_connect_timeout 10s;
proxy_read_timeout 600s;
proxy_buffering off;
error_page 502 503 = @router_fallback;
}
location @router_fallback {
proxy_pass http://router_api;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header Authorization $http_authorization;
proxy_read_timeout 600s;
}
location /admin/ {
@@ -49,70 +66,197 @@ http {
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header Authorization $http_authorization;
proxy_read_timeout 600s;
}
# SSE streaming endpoint
location /stream {
proxy_pass http://router_api;
proxy_pass http://router_api/stream;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header Connection "";
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection "upgrade";
proxy_buffering off;
chunked_transfer_encoding off;
}
# Dashboard API proxy for SSE
location /api/ {
proxy_pass http://dashboard_ui;
proxy_pass http://router_api/;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_buffering off;
}
# LiteLLM debug
location /litellm/ {
rewrite ^/litellm/(.*) /$1 break;
proxy_pass http://litellm_backend;
# LiteLLM gateway access (agents with new virtual keys)
location /litellm/v1/ {
proxy_pass http://litellm_backend/v1/;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header Authorization $http_authorization;
}
# Professional Dashboard (Phase 1-3) - Static HTML served via Nginx
location /dashboard/ {
alias /opt/inference-harness/dashboard/;
index dashboard.html;
add_header Cache-Control "public, max-age=3600";
add_header X-Content-Type-Options nosniff;
}
# Legacy Dashboard (root) - Proxy to Flask app
location / {
proxy_pass http://dashboard_ui;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_connect_timeout 10s;
proxy_read_timeout 600s;
proxy_buffering off;
}
# Performance analytics
# LiteLLM UI static assets
location /litellm-asset-prefix/ {
proxy_pass http://litellm_backend/litellm-asset-prefix/;
proxy_http_version 1.1;
proxy_set_header Host $host;
}
# LiteLLM Admin UI via /litellm/ prefix (LAN/internal access on :80)
location /litellm/ {
proxy_pass http://litellm_backend/;
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 X-Forwarded-Proto $scheme;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection "upgrade";
proxy_read_timeout 86400s;
proxy_buffering off;
proxy_redirect http://$host/ /litellm/;
proxy_redirect https://$host/ /litellm/;
}
location /dashboard/ {
proxy_pass http://dashboard_ui/;
proxy_http_version 1.1;
proxy_set_header Host $host;
}
location / {
root /opt/inference-harness/dashboard;
try_files $uri $uri/ /index.html;
}
location /metrics/ {
proxy_pass http://router_api;
proxy_pass http://router_api/metrics/;
proxy_http_version 1.1;
proxy_set_header Host $host;
}
# Circuit Breaker metrics (Phase 1)
location /metrics/circuit-breaker {
proxy_pass http://router_api/metrics/circuit-breaker;
proxy_http_version 1.1;
proxy_set_header Host $host;
}
location /router/ {
proxy_pass http://router_api/;
proxy_http_version 1.1;
proxy_set_header Host $host;
}
location /health/unified {
proxy_pass http://router_api/health/unified;
proxy_http_version 1.1;
proxy_set_header Host $host;
}
location /health {
proxy_pass http://router_api/health;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
}
}
# ════════════════════════════════════════════════════════════════
# Server :4000 — external LiteLLM entrypoint (cloudflared target)
# Fixes the /ui redirect: forces correct Host + scheme so LiteLLM
# builds external URLs instead of leaking 192.168.68.116:4000.
# LiteLLM itself is now bound to 127.0.0.1 only (see compose).
# ════════════════════════════════════════════════════════════════
server {
listen 4000;
# Canonical external identity — overrides whatever Host cloudflared sends
proxy_set_header Host litellm.sysloggh.net;
proxy_set_header X-Forwarded-Host litellm.sysloggh.net;
proxy_set_header X-Forwarded-Proto https;
proxy_set_header X-Forwarded-Port 443;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Real-IP $remote_addr;
# /ui → /ui/ — absolute redirect (cloudflared rewrites Host to origin IP,
# so a relative return would be absolutized to http://192.168.68.116:4000/).
location = /ui { return 301 https://litellm.sysloggh.net/ui/; }
# LiteLLM Admin UI + WebSocket
location /ui/ {
proxy_pass http://litellm_backend/ui/;
proxy_http_version 1.1;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection "upgrade";
proxy_read_timeout 86400s;
proxy_buffering off;
proxy_redirect http://litellm_backend/ https://litellm.sysloggh.net/;
proxy_redirect http://litellm.sysloggh.net:4000/ https://litellm.sysloggh.net/;
}
# UI static assets
location /litellm-asset-prefix/ {
proxy_pass http://litellm_backend/litellm-asset-prefix/;
proxy_http_version 1.1;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection "upgrade";
}
# SSO callback
location /sso/ {
proxy_pass http://litellm_backend/sso/;
proxy_http_version 1.1;
proxy_read_timeout 600s;
}
# API
location /v1/ {
proxy_pass http://litellm_backend/v1/;
proxy_http_version 1.1;
proxy_set_header Authorization $http_authorization;
proxy_read_timeout 600s;
proxy_buffering off;
}
# Key management + admin API
location /key/ {
proxy_pass http://litellm_backend/key/;
proxy_http_version 1.1;
proxy_set_header Authorization $http_authorization;
}
location /user/ {
proxy_pass http://litellm_backend/user/;
proxy_http_version 1.1;
proxy_set_header Authorization $http_authorization;
}
location /model/ {
proxy_pass http://litellm_backend/model/;
proxy_http_version 1.1;
proxy_set_header Authorization $http_authorization;
}
location /team/ {
proxy_pass http://litellm_backend/team/;
proxy_http_version 1.1;
proxy_set_header Authorization $http_authorization;
}
# Health
location /health {
proxy_pass http://litellm_backend/health;
proxy_http_version 1.1;
}
# Root + everything else → LiteLLM (UI index, /configs, /spend, etc.)
location / {
proxy_pass http://litellm_backend/;
proxy_http_version 1.1;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection "upgrade";
proxy_read_timeout 86400s;
proxy_buffering off;
proxy_redirect http://192.168.68.116:4000/ https://litellm.sysloggh.net/;
proxy_redirect https://192.168.68.116:4000/ https://litellm.sysloggh.net/;
}
}
}