Complete the transition to the new Dockerfile (#585)

Dockerfile

@@ -1,10 +1,14 @@
 FROM ghcr.io/astral-sh/uv:python3.11-bookworm
 
-COPY ./requirements-common.txt /requirements-common.txt
-COPY ./requirements-prod.txt /requirements-prod.txt
-RUN uv pip install --system -r /requirements-prod.txt
+ENV PYTHONDONTWRITEBYTECODE=1
+ENV PYTHONUNBUFFERED=1
+ENV UV_COMPILE_BYTECODE=1
 
-WORKDIR /app
+WORKDIR /hyperion
+
+COPY requirements-common.txt .
+COPY requirements-prod.txt .
+RUN uv pip install --system --no-cache-dir -r requirements-prod.txt
 
 COPY init.py .
 COPY alembic.ini .

README.md

@@ -183,7 +183,7 @@ curl --location --request POST 'http://127.0.0.1:8000/users/make-admin'
 
 Install docker and the compose plugin (https://docs.docker.com/compose/install/)
 
-`docker-compose.yaml` includes the minimal settings required to run Hyperion using docker compose. Docker images are based on [tiangolo/uvicorn-gunicorn-fastapi](https://github.com/tiangolo/uvicorn-gunicorn-fastapi-docker), and preloading the app is required to perform migrations and database initializations when running multiple workers.
+`docker-compose.yaml` includes the minimal settings required to run Hyperion using docker compose.
 
 > During dev, `docker-compose-dev.yaml` can be used to run the database, the redis server etc... If you really want to run the project without docker, you can do it but you will have to install the database, redis, etc ... yourself or disable corresponding features in the .env file (which is not recommended).
 
@@ -212,10 +212,8 @@ See [app/core/google_api/README.md](./app/core/google_api/README.md) for more in
 
 ---
 
-## Running Hyperion with Gunicorn
+## Hyperion deployment
 
-For production we encourage to use Gunicorn to run and manage multiple Uvicorn workers. You can use our [docker image](./Dockerfile) and [docker-compose file](./docker-compose.yaml) files to run Hyperion with Gunicorn. See [Gunicorn with Uvicorn](https://fastapi.tiangolo.com/deployment/server-workers/#gunicorn-with-uvicorn-workers) FastAPI documentation.
+For production we encourage to use multiple Uvicorn workers. You can use our [docker image](./Dockerfile) and [docker-compose file](./docker-compose.yaml) files to run Hyperion with Unicorn.
 
-Do not use gunicorn `--preload` flag. It initialise a first Hyperion instance then fork it to create workers. This is not compatible with the way we handle loggers in their own thread.
-
-You should use our [Gunicorn configuration file](./gunicorn_conf.py) to ensure that database initialization and migrations are only run once.
+You should use our [init file](./init.py) to ensure that database initialization and migrations are only run once.

app/core/config.py

@@ -200,7 +200,7 @@ class Settings(BaseSettings):
     ######################################
 
     # If Hyperion should initialize the database on startup
-    # This environment variable is set by the Gunicorn on_starting hook, to tell the workers to avoid initializing the database
+    # This environment variable is set by our init Python file to tell the workers to avoid initializing the database
     # You don't want to set this variable manually
     HYPERION_INIT_DB: bool = True
 

app/core/log.py

@@ -245,18 +245,6 @@ def get_config_dict(self, settings: Settings):
                     "level": MINIMUM_LOG_LEVEL,
                     "propagate": False,
                 },
-                # In production we use gunicorn instead of uvicorn
-                # We disable "uvicorn.access" to replace it with our custom "hyperion.access" which add custom information like the request_id
-                "gunicorn.access": {"handlers": []},
-                "gunicorn.error": {
-                    "handlers": [
-                        "file_errors",
-                        "matrix_errors",
-                        "console",
-                    ],
-                    "level": MINIMUM_LOG_LEVEL,
-                    "propagate": False,
-                },
             },
         }
 

app/main.py

@@ -1,8 +1,8 @@
-"""File used to start the application. It is used by uvicorn and gunicorn"""
+"""File used by Uvicorn to start the application."""
 
 from app.app import get_application
 from app.dependencies import get_settings
 
-# When using uvicorn or gunicorn, the application is started with the following command:
+# The application is started with the following function call:
 # We dissociate this step from the app.py file so that during tests we can initialize it with the mocked settings
 app = get_application(settings=get_settings())

init.py

@@ -1,5 +1,4 @@
 import logging
-import os
 
 from app.app import init_db
 from app.core.config import construct_prod_settings
@@ -10,10 +9,6 @@
 # - we will edit environment variables to avoid initializing the database and `get_settings()` is a cached function
 settings = construct_prod_settings()
 
-# We set an environment variable to tell workers to avoid initializing the database
-# as we want to do it only once before workers are forked from the arbiter
-os.environ["HYPERION_INIT_DB"] = "False"
-
 # Initialize loggers
 LogConfig().initialize_loggers(settings=settings)
 

start.sh

@@ -2,6 +2,10 @@
 set -e
 
 echo "Running migrations..."
+# We set an environment variable to tell workers to avoid initializing the database
+# as we want to do it only once before workers are created by Unicorn
+export HYPERION_INIT_DB="False"
 python3 init.py
+
 echo "Starting FastAPI server..."
 exec fastapi run app/main.py --workers ${NB_WORKERS:-1}