perf: Use Granian as app server

Much more performant and secure, made in Rust.

Author: clemlesne

Makefile

@@ -107,17 +107,15 @@ tunnel:
 	devtunnel host $(tunnel_name)
 
 dev:
-	VERSION=$(version_full) PUBLIC_DOMAIN=$(tunnel_url) uv run gunicorn app.main:api \
-		--access-logfile - \
-		--bind 0.0.0.0:8080 \
-		--graceful-timeout 60 \
-		--proxy-protocol \
+	VERSION=$(version_full) PUBLIC_DOMAIN=$(tunnel_url) uv run granian \
+		--host 0.0.0.0 \
+		--interface asgi \
+		--log-level info \
+		--port 8080 \
 		--reload \
-		--reload-extra-file .env \
-		--reload-extra-file config.yaml \
-		--timeout 60 \
-		--worker-class uvicorn.workers.UvicornWorker \
-		--workers 2
+		--workers 2 \
+		--workers-kill-timeout 60 \
+		app.main:api
 
 build:
 	DOCKER_BUILDKIT=1 docker build \

cicd/Dockerfile

@@ -33,4 +33,4 @@ ARG VERSION
 ENV VERSION=${VERSION}
 
 # Starting the backend
-CMD gunicorn app.main:api --bind 0.0.0.0:8080 --graceful-timeout 60 --proxy-protocol --timeout 60 --worker-class uvicorn.workers.UvicornWorker --workers 4
+CMD ["granian", "--interface", "asgi", "--host", "0.0.0.0", "--port", "8080", "--log-level", "critical", "--workers-kill-timeout", "60", "--workers", "4", "app.main:api"]

pyproject.toml

@@ -23,7 +23,7 @@ dependencies = [
   "azure-storage-queue~=12.12",                          # Azure Storage Queue
   "django-htmlmin~=0.11",                                # Minify HTML
   "fastapi~=0.115",                                      # Web framework
-  "gunicorn~=23.0",                                      # Application server
+  "granian~=2.3",                                        # Application server
   "jinja2~=3.1",                                         # Template engine, used for prompts and web views
   "json-repair~=0.30",                                   # Repair JSON files from LLM
   "mistune~=3.0",                                        # Markdown parser for web views
@@ -46,13 +46,13 @@ dependencies = [
   "tiktoken~=0.8",                                       # Tokenization library for OpenAI models
   "twilio~=9.3",                                         # Twilio SDK, used for SMS
   "typing-extensions~=4.12",                             # Typing extensions for Python 3.6+
-  "uvicorn[standard]~=0.32",                             # Application middleware
 ]
 
 [project.optional-dependencies]
 dev = [
   "deepeval~=0.21",            # LLM model evaluation
   "deptry~=0.20",              # Dependency tree testing
+  "granian[reload]",           # Granian with reload support for development
   "pyright~=1.1",              # Static type checker
   "pytest-assume~=2.4",        # Pytest plugin for conditional tests
   "pytest-asyncio~=0.24",      # Pytest plugin for async tests