feat(metrics): add lua script for prometheus metrics on latency

Author: Frederic Spiers

docker/nginx/nginx.conf

@@ -1,4 +1,5 @@
 load_module "/usr/lib/nginx/modules/ngx_stream_module.so";
+load_module "/usr/lib/nginx/modules/ngx_lua_module.so";
 
 worker_processes 1;
 
@@ -26,6 +27,13 @@ http {
             return 404;
         }
 
+        location /metrics {
+            content_by_lua_block {
+                local prometheus = require "resty.prometheus"
+                prometheus:collect()
+            }
+        }
+
         location /healthz {
             default_type text/plain;
             return 200 "OK\n";
@@ -34,6 +42,7 @@ http {
 }
 
 stream {
+    lua_shared_dict prometheus_metrics 10M;
     map_hash_bucket_size 128;
     map_hash_max_size 4096;
 
@@ -47,10 +56,20 @@ stream {
     resolver kube-dns.kube-system.svc.cluster.local valid=30s;
     resolver_timeout 5s;
 
+    init_by_lua_block {
+        local prom = require "docker.nginx.prometheus"
+        prom.init()
+    }
+
     server {
         listen 127.0.0.1:8443 proxy_protocol;
         ssl_preread on;
         proxy_pass $ssl_preread_server_name:443;
         proxy_protocol off;
+
+        log_by_lua_block {
+            local prom = require "docker.nginx.prometheus"
+            prom.log_connect_time()
+        }
     }
 }

docker/nginx/prometheus.lua

@@ -0,0 +1,24 @@
+  local prometheus = require "resty.prometheus"
+  local metric_connections, metric_connect_time
+
+  local function init()
+      prometheus = prometheus.init("prometheus_metrics")
+      metric_connections = prometheus:counter("nginx_stream_connections_total", "Total connections", {"upstream"})
+      metric_connect_time = prometheus:histogram("nginx_stream_upstream_connect_seconds", "Upstream connect time",
+  {"upstream"})
+  end
+
+  local function log_connect_time()
+      local connect_time = tonumber(ngx.var.upstream_connect_time)
+      local upstream = ngx.var.upstream_addr or "unknown"
+
+      if connect_time then
+          metric_connect_time:observe(connect_time / 1000, {upstream})  -- Convert ms to seconds
+          metric_connections:inc(1, {upstream})
+      end
+  end
+
+  return {
+      init = init,
+      log_connect_time = log_connect_time
+  }

melange/ggbridge.yaml

@@ -10,6 +10,8 @@ package:
     runtime:
       - nginx
       - nginx-mod-stream
+      - nginx-mod-lua
+      - lua5.1-resty-core
       - wstunnel
   target-architecture:
     - aarch64
@@ -39,4 +41,5 @@ pipeline:
       install -dm755 "${{targets.destdir}}"/etc/ggbridge/tls
       install -m755 docker/scripts/run.sh "${{targets.destdir}}"/opt/ggbridge/run.sh
       install -m644 docker/nginx/nginx.conf "${{targets.destdir}}"/etc/ggbridge/nginx.conf
+      install -m644 docker/nginx/prometheus.lua "${{targets.destdir}}"/etc/ggbridge/prometheus.lua
   - uses: strip