liberty-bikes-ai

Objective

In this lab you will write your own microservice for Liberty Bikes that uses OpenLiberty, Jakarta EE 8, and MicroProfile 3. Liberty Bikes is a real-time web-based multiplayer game that runs on Liberty and demonstrates a microservice architecture as well as the previously mentioned technologies.

Normally Liberty Bikes is played by a human looking at the screen and manually controlling the direction of their character. However, in this lab you will be coding your own AI microservice that will read the game board and make direction changes automatically. You will spend most of the lab running all of the services locally to build and test your AI service. At the end of the lab, the lab administrators will run the core Liberty Bikes services and the lab attendees (you) will run just your AI service and register it with the core Liberty Bikes services being hosted by the lab admins.

0) Prereqs

Required software:

• Have Java 8 or 11 installed
• Have git installed
• Your preferred code editor

Fork and clone the liberty-bikes repository at: https://github.com/OpenLiberty/liberty-bikes

Start up the Liberty Bikes core services by running the commands:

cd liberty-bikes
./gradlew start
./gradlew stop

This will start and stop what we will refer to as the "core services".  The commands will also download a bunch of things such as gradle itself, maven dependencies, and the Liberty runtime. All of the command should complete successfully. If they do not, verify you have all of the prereqs software installed and then ask one of the lab admins for assistance.

Fork and clone the liberty-bikes-ai repository at: https://github.com/OpenLiberty/liberty-bikes-ai

If you get stuck

If you are stuck on a section and unable to make progress, please ask one of the lab admins for help. As a backup, we have tagged several checkpoints along the lab. If you get stuck and just want to skip to the next section, you can do the following: If you have registered your Bot key in src/main/resources/META-INF/microprofile-config.properties at this point, save it off to a file outside the lab repo Wipe away all of your current changes with git checkout -- . Check out the next checkpoint branch with git checkout checkpoint-1 (or whatever checkpoint is next, checkpoint names are labelled in section headers) Restart your bot server by doing ./gradlew start

1) Your first game

Before you start writing your AI service, let's begin by getting a basic understanding of what Liberty Bikes is on the surface and how it all works under the hood.

How the game works:

• The game consists of rounds that typically last 20-60 seconds each
• Each round starts with up to 4 players in opposite corners of the map
• When the starting countdown reaches 0, players begin moving in whatever direction they are facing
• Players may use arrow keys (up/down/left/right) to change directions
• If a player runs into anything (a wall, the map border, a moving obstacle, another player's trail, or their own trail) they die
• The round continues until there is only 1 player alive
• The last player alive gets 1st place, the second to last player gets 2nd place, and so on. Points are awarded (or subtracted) based on what place you came in

Running the core services:

The stuff in the main "liberty-bikes" repository contains the core services of Liberty Bikes. To start all of the core services on your laptop, run the ./gradlew start command from the root of the liberty-bikes repository.

To play your first game:

1. start all of the core services with ./gradlew start if you haven't already
2. open the Liberty Bikes web UI by running the ./gradlew frontend:open command, or open a web browser to http://localhost:12000 NOTE: It may be helpful to bookmark http://localhost:12000 for the duration of this lab
3. Press "Play as Guest"
4. Enter a username for your character, then "Sign in as Guest"
5. Press "play now". You should see a game board on the left and player status and scores on the right.
6. Use the arrow keys on your keyboard to change directions (if nothing is happening, make sure you have the web browser window in focus/selected)
7. Requeue into another game by pressing the "requeue" button at the bottom right of the screen after the round is over
8. After you requeue, you should see your stats updated on the right hand side
9. Close the browser tab when you are done

To host your first game:

1. open a new browser tab, and go to the liberty bikes web ui
2. If you are logged in, log out
3. Press "Host round". You will see a game screen with 0 players in it and a timer countdown running
4. Open a new browser tab and go to the liberty bikes web ui
5. Sign in as a guest again. If you pressed "play now" while a round was in progress, you will be entered into the queue. If a round was not in progress, you will be entered into the next round.
6. Close out both browser tabs

How it works:

The Liberty Bikes core services are comprised of 4 OpenLiberty servers, each running 1 microservice:

• frontend: Runs the Angular-based UI (features used: Servlet)
• game-service: Runs the core game engine and coordinates rounds. (features used: CDI, EE Concurrency, JAX-RS, JSON-B, MP Config, MP Rest Client, MP JWT, Websockets)
• player-service: Manages player registration and stats. (features used: CDI, JAX-RS, JSON-B, MP Config, JDBC)
• auth-service: Coordinates authentication and distributes JWTs (JSON Web Tokens) for other services. Can be set up to integrate with Google, Twitter, and Github SSO providers, but that will not be covered in this lab.

To see technology highlights in the code, have a look at https://github.com/OpenLiberty/liberty-bikes#technologies-used, or just poke through the code (particularly in game-service and player-service).

2) Hello World!

NOTE: For the remainder of this lab (except for section 5) you can leave the core services running on your machine. From this point forward you will be developing an app that runs in the 'liberty-bikes-ai' github repo that you cloned in section 0.

Import the project

Using some sort of IDE (Eclipse, IntelliJ, or VS Code), import the 'liberty-bikes-ai' folder as an existing Gradle project.  Ensure that you have build automatically turned on, or remember to issue a rebuild of the project whenever you make a code change.

Start your AI server and see the logs

In the root of the 'liberty-bikes-ai' repo, run ./gradlew start to start up your AI server. Next, open the server logs by going to build/wlp/usr/servers/ai-service/logs/messages.log. You will see content similar to this:

[AUDIT   ] CWWKE0001I: The server ai-service has been launched.
[AUDIT   ] CWWKZ0058I: Monitoring dropins for applications.
[AUDIT   ] CWWKT0016I: Web application available (default_host): http://localhost:8083/liberty-bikes-ai/
[AUDIT   ] CWWKZ0001I: Application liberty-bikes-ai started in 0.462 seconds.
[AUDIT   ] CWWKF0012I: The server installed the following features: [cdi-2.0, jaxrs-2.1, jaxrsClient-2.1, jndi-1.0, jsonb-1.0, jsonp-1.1, mpConfig-1.3, mpRestClient-1.3, servlet-4.0, websocket-1.1].
[AUDIT   ] CWWKF0011I: The ai-service server is ready to run a smarter planet. The ai-service server started in 1.385 seconds.

Leave this log file open for duration of this lab -- you will be referring to it frequently to see the output of your application.

Hot code updates with OpenLiberty

A very nice feature of OpenLiberty is that for local development you don't need to restart your server or even redeploy your app between code changes. If you are using an IDE that compiles the app for you, the IDE recompiling the app will be detected by Liberty and it will automatically restart the app with the new code. To test this out, go to the org.libertybikes.ai.service.StartupProcedure class and make any code change you like (as long as it compiles). Once you save the code change and your IDE recompiles the app, refresh your messages.log file and you will see the app updated automatically in a few milliseconds:

[AUDIT   ] CWWKT0017I: Web application removed (default_host): http://localhost:8083/liberty-bikes-ai/
[AUDIT   ] CWWKZ0009I: The application liberty-bikes-ai has stopped successfully.
[AUDIT   ] CWWKT0016I: Web application available (default_host): http://localhost:8083/liberty-bikes-ai/
[AUDIT   ] CWWKZ0003I: The application liberty-bikes-ai updated in 0.152 seconds.

Starting the registration procedure at application startup time

When we start the server with our AI bot application, we don't want to have to manually initiate the registration procedure with the core services. Lets write a web listener that will automatically be invoked when the application is started. Update the StartupProcedure class to be the following:

@WebListener
public class StartupProcedure implements ServletContextListener {

    @PostConstruct
    public void joinFirstRound() {
        System.out.println("Hello world!");
    }
}

When you save this code change, the app should restart again and you should see the "Hello world!" message in your console.log.

Implementing the registration procedure [checkpoint-1]

First, update StartupProcedure to @Inject a RegistrationBean and call joinRound(). When you save the change, you should see print statements in your logs indicating joinRound() was invoked.

In order to register our AI Bot with the core services so it can play in games, we need to do two things: get the current party ID from the core services join the party queue so our bot waits its turn to play games, and can be notified when it's time to join a game

We can get the current party ID by invoking GET http://<game-service-ip-port>/party/describe.  The response will be a JSON with the format:

{ "partyId" : String }

Making REST calls with MicroProfile REST Client

Go to the GameServiceClient interface and add an interface method that represents the endpoint we want to invoke like so:

    @GET    
    @Path("/party/describe")
    @Produces(MediaType.APPLICATION_JSON)
    public Map<String, Object> describe();

If you are familiar with JAX-RS, you will notice that this is the exact same signature and annotations that a JAX-RS method would have, but without the method body. Next, we need to declare the base URI of this REST client. We can do this using the MP REST Client annotations on the interface:

@Path("/")
@Produces(MediaType.APPLICATION_JSON)
@RegisterRestClient(baseUri = "http://localhost:8080", configKey = AIConfiguration.GAME_SERVICE_KEY)
public interface GameServiceClient {

When we run the game-service locally, it's URI is http://localhost:8080, so we will default to using that. The configKey attribute allows us to override the baseUri using MP Config, which we will do later on in section 5.

Now that we have our REST Client bean, we can inject it into the RegistrationBean and get the current round id:

    // don't forget to add these 2 annotations!
    @Inject
    @RestClient
    GameServiceClient gameService;


    public void joinRound() {
        System.out.println("Attempting to register with game service...");


        // TODO: use MPRestClient to pull the valid partyID from the game service
        String partyId = (String) gameService.describe().get("partyId");
        System.out.println("Found party id: " + partyId);

After building this change you should see that a 4-character code was received like this: Found party id: WDPW, followed by a NPE because we haven't implemented the subsequent code yet.

Re-queueing with JAX-RS Server Sent Events

Now that we know the party ID, we need to register our bot in the party queue. We can use JAX-RS Server Sent Events (SSE) to put ourselves into the queue for the party. Since only 4 people can play in a game at a time, this will ensure everyone waits their turn. By establishing an SSE request at http://<game-service-ip-port>/party/{partyId}/queue?playerId={playerId} the core services will put our bot in line for a game, and then asynchronously notify us when it's our turn to join a round.

In order to make this request, we need to know 1 more piece of information: our Bot player's ID. To get this, we need to go to the frontend for the core services we want to register with. Right now, that's http://localhost:12000/ (later it will be one of the lab admins IP address). Click the "Register Bot" button, enter a name for your bot, and then click "Register Bot" again to get your bot player's ID. Copy and paste the ID string into src/main/resources/META-INF/microprofile-config.properties as the value for player_key.

In the AIConfiguration class, add @Inject and @ConfigProperty(name = "player_key") to the playerKey field. This will cause the value from microprofile-config.properties to be injected into the field, which is nice because we don't need to add secret information to our codebase, and we can easily change or override the value later without needing to recompile or repackage the app later.

We also need to add the game-service URL to the AIConfiguration class, add the following annotations to the gameServiceUrl field:

    @Inject
    @ConfigProperty(name = GAME_SERVICE_KEY + "/mp-rest/url", defaultValue = "http://localhost:8080")
    private String gameServiceUrl;

To access the AIConfiguration from the RegistrationBean, add the @Inject annotation to the RegistrationBean.config field.

Now that our player ID is configured, we can make the JAX-RS SSE request. Add the following code to the end of the RegistrationBean.joinRound() method:

        SseEventSource source = SseEventSource.target(target).build();  
        source.register(inboundEvent -> processInboundEvent(inboundEvent, source),      
                        errEvent -> errEvent.printStackTrace(), 
                        () -> System.out.println("Closing SSE source for " + targetStr));       
        source.open();

This code will make an HTTP request to the game-server with the URL discussed above. Here our AI Bot app is acting as the client to the SSE request, so it doesn't need to constantly poll the game-service to see what position in the queue it is -- the server will send the event when it wants to update clients. Most of the magic happens in the SseEventSource.register() method, which will set up callbacks for inbound events, error events, and the close event respectively.

Verify your progress

At this point, save any code changes you've made to trigger an app update. If you check your messages.log you should see some JSON messages coming through from the game-server. Also, you can go to http://localhost:12000 and select "host game". You should see your bot player in the game! It will just go straight and crash and probably get last place, but we will make it smart later. When the game is over your host view should automatically requeue to the next round, and so should your bot. 

3) Lets get moving [checkpoint-2]

At this point your AI bot will automatically join games when your application starts, and it when a game is over it will move itself to the next round. But, it's not a very good player yet. The next step we need to do is try to make some moves on the game board.

If you check your server logs, you will see a bunch of messages like got json: <bunch of json data>. This data is being sent from the game-service every "tick" (50ms increments of time) of the game, which represents the state of the game board. We can use this data to render our own understanding of the current game board, and try to make decisions on which way our bot should move.

Sending and receiving messages with Websockets

First, we will cover making our bot change directions. If you look at the AIWebsocket class you will see a few annotated methods: onOpen(), onMessage(), and onClose(), which are invoked whenever the websocket is opened, receives a message, or closed, respectively. The onMessage() method only does one thing right now, and that is initiating the requeue procedure whenever a FINISHED message is received from the game-service. In the else block of that method, we can respond to game ticks using the sendDirection(DIRECTION) method, which will send a response over the websocket indicating which way our Bot should go. 

To make your Bot change directions, add a simple tick counter (i.e. count the number of times onMessage() is called) and send a message DOWN once every 10 ticks, otherwise go RIGHT. To be more efficient, track which direction your Bot is currently facing and only call sendDirection() if your Bot's direction has changed.

Verify your progress

Save your changes to trigger an app update. You should see your Bot changing directions about once every 500ms.

Modelling JSON data with JSON-B

Now we will make sense of the JSON data coming from the game-service using JSON-B. There are 3 empty classes in the org.libertybikes.ai.model package for you to fill out. Each class has the JSON schema documented in it. JSON-B allows us to model JSON data with Java objects, an easy convert back and forth between JSON and POJO (plain old java object). For example, suppose we have the following JSON data representing a dog:

{ 
  "name": "Buddy",
  "age": 5,
  "bites": false
  "favoriteToys": [ "stick", "tennis ball", "bone" ]
}

We can model this JSON data as a Java class like this:

public class Dog {
  public String name;
  public int age;
  public boolean bites;
  public List<String> favoriteToys;
}

Then, we can convert from JSON data <--> POJO like this:

String json = // ...
Dog buddy = jsonb.fromJson(json, Dog.class);
String afterJson = jsonb.toJson(buddy);

Some other relevant aspects of JSON-B are:

• If an unrecognized property is found in the JSON data (i.e. it has no matching Java property) it is ignored
• By default, java fields that did not get matched to incoming JSON data are left as null
• By default, JSON-B only maps to public fields or getter/setter methods. Non-public members are ignored
• By default, the java objects must have a default (no-args) constructor.

Using this information, try to model the JSON data as POJOs in your app.  If you get stuck, there is more information at: http://json-b.net/docs/user-guide.html#default-mapping.  Or if you really get stuck, the answers can be found here: https://github.com/OpenLiberty/liberty-bikes-ai/tree/checkpoint-3/src/main/java/org/libertybikes/ai/model

4) Code, Play, Repeat

Now that you've modeled your JSON data with Java objects, you can pass these objects to your AILogic and finally make some informed decisions on where to steer your bot! Be warned, there are a few different types of messages that the game-service will broadcast -- not all of them are game ticks! You will need to account for this when passing a GameTick to your AILogic class somehow:

            } else {
                // use jsonb to convert from String --> POJO
                GameTick gameTick = jsonb.fromJson(message, GameTick.class);
                if (gameTick.isValid())
                    sendDirection(aiLogic.processAiMove(gameTick));
            }

The AILogic class can house any of the... AI logic... that you want to use to control your bot. Feel free to create additional methods and classes at this point if it would be helpful -- this is essentially the end of the lab and you can do put as much or as little effort as you'd like into making your AI Bot. Perhaps think of some algorithms that you could use to steer your bot. For example, here are a few basic algorithms:

• Go straight until you are about to run into something, then turn left or right (relative to bot's current direction)
• At each game tick "look" in each direction -- forward, left, and right. Compute which direction you can go in a straight line for the longest, and go that way.

5) WIP