From 3f5c8562dde3b3e44114f7a75c2b9c5ab3ddd6c6 Mon Sep 17 00:00:00 2001
From: "John A. Hoeven" <git@cervelloelettrico.it>
Date: Mon, 20 Jul 2026 00:20:44 +0200
Subject: [PATCH] =?UTF-8?q?Add=20tracked=20nginx=20reverse-proxy=20templat?=
 =?UTF-8?q?e=20for=20llama-server=20=E2=80=94=20previously=20only=20existe?=
 =?UTF-8?q?d=20as=20a=20heredoc=20run=20live=20on=20BigBoy;=20also=20resol?=
 =?UTF-8?q?ves=20the=20server=5Fname=20wildcard=20conflict=20warning?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 ansible/templates/llama.conf.j2 | 39 +++++++++++++++++++++++++++++++++
 1 file changed, 39 insertions(+)
 create mode 100644 ansible/templates/llama.conf.j2

diff --git a/ansible/templates/llama.conf.j2 b/ansible/templates/llama.conf.j2
new file mode 100644
index 0000000..41470a3
--- /dev/null
+++ b/ansible/templates/llama.conf.j2
@@ -0,0 +1,39 @@
+# llama.conf — nginx reverse proxy for BigBoy's llama-server
+#
+# This is a Jinja2 template (Ansible's `template` module convention),
+# ready to use once Phase 5 is formalized into an actual Ansible role.
+# Until then, it's the tracked, reproducible reference for what's
+# actually configured live on BigBoy — this previously existed only as
+# a heredoc typed directly during the deployment session, never
+# committed anywhere.
+#
+# Manual deployment (until this is wired into a real Ansible role):
+#   1. Fill in the Jinja2 variables below with their actual values
+#   2. Copy the rendered result to /etc/nginx/conf.d/llama.conf
+#   3. sudo nginx -t   (confirm syntax before reloading)
+#   4. sudo systemctl reload nginx
+#
+# NOTE on server_name: the first live deployment used `server_name _;`
+# (the catch-all wildcard) and hit a harmless-but-noisy warning —
+# "conflicting server name "_" on 0.0.0.0:80, ignored" — because
+# AlmaLinux's stock nginx package already ships its own default catch-all
+# server block using the same wildcard. This template uses the actual
+# host IP instead, avoiding the duplicate entirely rather than just
+# tolerating the warning.
+
+server {
+    listen 80;
+    server_name {{ ip_address }};
+
+    location / {
+        proxy_pass http://{{ llama_server_host }}:{{ llama_server_port }};
+
+        # Required for llama-server's streaming (server-sent event)
+        # responses — without these, chat responses arrive all at once
+        # instead of streaming token-by-token.
+        proxy_http_version 1.1;
+        proxy_set_header Connection "";
+        proxy_buffering off;
+        proxy_read_timeout 300s;
+    }
+}
\ No newline at end of file