MultiJSnake 🐍🐍🐍

Snake played on a server -- communication happens mostly via Websockets. [https://multijsnake-prod-main-0wuwfu.mo4.mogenius.io](Play it in your browser). Either alone, with friends (via an invite link) or against computer players.

It also offers a static stand-alone mode to watch different neural net based autopilots at snake.schawe.me. Yes, the neural nets are executed in your browser.

⚙️ Setup

You just need Java 11 to be installed on your system, other dependencies (Maven, NPM, ...) will be fetched by the build script. So just run

./mvnw spring-boot:run

And the server will be compiled and started on http://localhost:8080.

🤖 Training an autopilot

You can train a neural net to play the game with the scripts in src/main/py. As prerequisites there must be Python 3.6+ available and the dependencies from requirements.txt need to be installed, e.g., with pip install -r requirements.txt.

Also ensure that the java classes are are compiled (e.g. by executing the above step,) since the training will need them. More information about this curiosity can be found at blog.codecentric.de/2021/11/java-klassen-python.

The training can be started and continued by changing into src/main/py/ and executing the train_AC.py and train_convAC.py scripts.

For converting the file with the trained model into a format suitable for tensorflowjs use:

cp yourModelFile.keras strippedModelFile.keras
# remove some data unecessary for inference (but important for training)
python3 strippedModelFile.keras
tensorflowjs_converter --input_format=keras strippedModelFile.keras tfjs_model

🧪 Tests

There are unit tests of the backend and end-to-end tests using cypress. You can run them via

./mvnw verify

🛠️ Deployment

Here are some options of how to deploy this server.

💻 Local (for development and testing)

Thanks to SpringBoot, this can be accomplished simply by executing

./mvnw spring-boot:run

No database setup is required (it is using an in-memory database).

🔌 Static

We can build a static version, which does not include any features needing the Java-Server, which can be deployed on any web space (e.g. GitHub pages or Netlify).

Just execute

bash buildStatic.sh

and copy the generated folder tmp/resources/static to the web, or serve it locally with

python3 -m http.server 8080 -d tmp/resources/static

Note that this does only include the AI demo and not the actual playable multiplayer game.

☁️ On Mogenius (or other)

Heroku will automatically use the correct build pack to start and execute the server.

Mogenius will automatically detect the Dockerfile and run.

However, for data persistency a Postgres database has to be created (possibly via a third party, eg. ElephantSQL) and the following environment variables should be set:

# Do not store the native libraries for all platforms, but only the relevant platform (otherwise the artifact is too large for Heroku)
MAVEN_CUSTOM_OPTS=-DskipTests=true -Djavacpp.platform=linux-x86_64
# configure the Postgres DB, these variables are all that is needed to switch from H2 to Postgres
# placeholders for secret values are marked here with []
SPRING_DATASOURCE_DRIVER-CLASS-NAME=org.postgresql.Driver
SPRING_DATASOURCE_PASSWORD=[password]
SPRING_DATASOURCE_TYPE=org.apache.tomcat.jdbc.pool.DataSource
SPRING_DATASOURCE_URL=jdbc:postgresql://[server]:[port]/[database]?sslmode=require
SPRING_DATASOURCE_USERNAME=[user]
SPRING_JPA_DATABASE-PLATFORM=org.hibernate.dialect.PostgreSQLDialect
SPRING_JPA_HIBERNATE_DDL-AUTO=update

🐳 With Docker/docker-compose

There is also a Dockerfile in this repository, which builds and executes the server on port 8080. This is used in the docker-compose.yml, which uses nginx as a reverse proxy and letsencrypt to serve over https.

To use it, ensure that the following environment variables needed for letsencrypt are set (it also works if they are saved in the file .env.)

DOMAIN=[domain]
EMAIL=[email]

Then the containers can be built and started with:

docker-compose build
docker-compose up