deploy/ecosystem/nginx.conf

@@ -31,49 +31,24 @@
 #     read the note at the top of that file first.
 #
 # `nginx -t` catches the second and not the first. Run it anyway, every time.
-
-# maven.<domain> → mavweb (docker-compose.yml publishes it on 127.0.0.1:9201).
-# Same bind + ACL as the siblings, and for a stronger reason: mavweb serves
-# POST /tools, which defines argv that internal/tool EXECUTES, plus POST
-# /routines, /api/revert and /api/chat (Vikunja #317). Without
-# -webauthn-origin/-webauthn-rpid mavweb has no auth of its own, so this block
-# is the auth. If you add TLS and a basic-auth/oauth2-proxy layer, keep the
-# allow/deny anyway — belt and braces on an RCE surface.
 #
-# WebSocket upgrade matters here: /ws carries push-to-talk audio, so the
-# Upgrade/Connection headers below are required, not decoration. They reference
-# $connection_upgrade, which this file does NOT define — see
-# nginx-upgrade-map.conf and point 2 above.
+# BLOCK ORDER IS LOAD-BEARING. nginx serves the first block for a given listen
+# address when no server_name matches, so whichever block comes first here
+# answers requests with an unknown or absent Host header. That must not be
+# mavweb: it is the one surface in this file that can define and run argv. The
+# nexus block is marked default_server so the choice is explicit rather than a
+# consequence of file order, and the maven block sits last as a second guard.
+# If you add a block, do not put it above nexus. If another site file already
+# claims default_server on 10.42.0.1:80 or 192.168.1.104:80, nginx refuses to
+# start with "a duplicate default server" — drop the two keywords here and rely
+# on the block order instead.
+#
+# Every server_name below is a literal for kvmx.ru even though the comments
+# write maven.<domain>. This file reads like a template and is not one.
 
 server {
-    listen 10.42.0.1:80;
-    listen 192.168.1.104:80;
-    server_name maven.kvmx.ru;
-
-    allow 10.42.0.0/24;
-    allow 192.168.1.0/24;
-    deny all;
-
-    # push-to-talk uploads raw PCM; the default 1m is enough for a short
-    # utterance but not for a long one.
-    client_max_body_size 32m;
-
-    location / {
-        proxy_pass http://127.0.0.1:9201;
-        proxy_http_version 1.1;
-        proxy_set_header Upgrade $http_upgrade;
-        proxy_set_header Connection $connection_upgrade;
-        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_read_timeout 300s;   # an LLM turn can take minutes on the iGPU
-    }
-}
-
-server {
-    listen 10.42.0.1:80;
-    listen 192.168.1.104:80;
+    listen 10.42.0.1:80 default_server;
+    listen 192.168.1.104:80 default_server;
     server_name nexus.kvmx.ru;
 
     allow 10.42.0.0/24;
@@ -124,3 +99,47 @@ server {
         proxy_set_header X-Forwarded-Proto $scheme;
     }
 }
+
+# maven.<domain> → mavweb (docker-compose.yml publishes it on 127.0.0.1:9201).
+# Same bind + ACL as the siblings, and for a stronger reason: mavweb serves
+# POST /tools, which defines argv that internal/tool EXECUTES, plus POST
+# /routines, /api/revert, /api/chat, /api/ptt and /ws (Vikunja #317). Without
+# -webauthn-origin/-webauthn-rpid mavweb has no auth of its own, so this block
+# is the auth. If you add TLS and a basic-auth/oauth2-proxy layer, keep the
+# allow/deny anyway — belt and braces on an RCE surface.
+#
+# WebSocket upgrade matters here: /ws carries push-to-talk audio, so the
+# Upgrade/Connection headers below are required, not decoration. They reference
+# $connection_upgrade, which this file does NOT define — see
+# nginx-upgrade-map.conf and point 2 above.
+
+server {
+    listen 10.42.0.1:80;
+    listen 192.168.1.104:80;
+    server_name maven.kvmx.ru;
+
+    allow 10.42.0.0/24;
+    allow 192.168.1.0/24;
+    deny all;
+
+    # push-to-talk uploads raw PCM; the default 1m is enough for a short
+    # utterance but not for a long one.
+    client_max_body_size 32m;
+    # A 32m upload over a slow link outlives the 60s default on the two body
+    # timeouts, and nginx cuts it at exactly 60s with a 408 or a 504 that looks
+    # like the turn failed. Raise them with the size, not just proxy_read_timeout.
+    client_body_timeout 300s;
+
+    location / {
+        proxy_pass http://127.0.0.1:9201;
+        proxy_http_version 1.1;
+        proxy_set_header Upgrade $http_upgrade;
+        proxy_set_header Connection $connection_upgrade;
+        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_send_timeout 300s;   # pushing the PCM upstream, same reason
+        proxy_read_timeout 300s;   # an LLM turn can take minutes on the iGPU
+    }
+}