GAMEON

Hands-on experiment building microservices and cloud native applications

JSR-107 Caching

Where we learn about Caching, the standards-based way.

Overview

In this adventure, we’ll learn about Caching, and will walk through adding Caching support via JSR-107 annotations to a Java-based Game On room.

By the end, we hope you’ll have an understanding of the value and use that Caching technologies bring to Cloud Native applications and microservices. We’ll leave you with some suggestions for further improvements to your room, so you can continue to explore the concepts.

Why Caching? Why JSR-107?

Caching is one of those awkward bits of function you can totally avoid adding when first creating a bit of code. Everything will work just fine during your initial testing, but you worry about what will happen as the usage begins to scale up.

Maybe you are trying to avoid invoking a remote service too frequently, maybe you just want to avoid incurring the cost of redoing a calculation.

At least for me, the chain of thoughts usually runs something like this;

"I should add a Cache, right here! I can just use some variant of a Map, but I’ll need to consider how items will ever leave the Cache. And what about concurrency ? performance ? testing ?".

It’s usually somewhere around there it that it dawns on me that I should probably look at how other people have solved this, as there’s probably a library I could use.

There are many Caching libraries for Java, ranging from simple in memory thread safe caches, to distributed transactional remote based services. And they’ve been around long enough that there’s been an effort to try to standardise an approach for them since way back in 2001. JSR-107, (or 'JCache') has been working toward providing a standard for Caching for almost 2 decades, and there are quite a few libraries out there that implement it.

For this walkthrough, we’ll be using Redisson, a library that provides a JSR-107 interface to a Redis server. Although Redisson provides a very capable API to talk to Redis, in this walkthrough we’ll be limiting ourselves to just the JSR-107 aspects, and showing how they can be used within a Game On room.

Hopefully we’ll be able to revist the Redisson API in a future walkthrough.

Prerequisites

This walkthrough will start with the default Java Sample Room. It assumes you have the Java Sample Room up and running as a cf app in Bluemix.

You will need to create a Redis instance in Bluemix, and associate it to your Java Sample Room app.

  1. Start by heading to the Bluemix Catalog, and find the Redis Cloud Service (under Data & Analytics)
  2. Scroll down through the Pricing Plans for Redis Cloud, and select the "30MB 1 Dedicated database" - "Free" option.
  3. In the Connect to: drop down on the left of the page, select entry for your Java Sample Room.
  4. Hit the Create button at the bottom right of the page.

Walkthrough

Adding JSR-107 to your room.

We want to use Redisson to provide our JSR-107 support, but that won’t get us the annotations (which are kinda cool). The annotation support is expected to come from the runtime, in our case that would be the Liberty CDI support, except that doesn’t have JSR-107 support today because JSR-107 isn’t part of the level of JEE it supports.

In the interim, the JSR-107 RI ships a a set of modules that can enable use of the annotations within CDI (and Spring, and Guice).

We could use those modules as-is, and with the right config file Redisson would know how to access our Redis, and we’d be just fine.

But, we’d rather not have to create that config file, as our Redis configuration information is sitting in the VCAP_SERVICES environment variable, and we’d like to use that.

To make things a little easier, we’ve prepared a fork of the CDI module, which allows the CacheManager used by the annotations to be supplied by the application code.

Adding the dependencies.

We’ll start by adding this special CDI module to our Java Sample Room as a library.

Firstly, edit the pom.xml in your room project and find the <dependencies>…​</dependencies> block. Add these dependencies after the existing ones, just before the </dependencies> tag.

<dependency>
    <groupId>org.redisson</groupId>
    <artifactId>redisson</artifactId>
    <version>3.2.3</version>
</dependency>
<dependency>
    <groupId>javax.cache</groupId>
    <artifactId>cache-api</artifactId>
    <version>1.0.0</version>
</dependency>
<dependency>
    <groupId>com.github.BarDweller</groupId>
    <artifactId>JSR107-RI-CDI-Custom-CacheManager</artifactId>
    <version>v1.0.9-STILETTO</version>
</dependency>

The first is the Redisson client, that will provide the implementation of the API for our room, the second provides the API interfaces, and the last is the CDI Module to enable the JSR-107 annotations.

Creating the default cache manager provider

The CDI Module allows us to configure the CacheManager the JSR-107 annotations should use. It provides this capability by using a Java Service, our room needs to include an implementation of the DefaultCacheManagerProvider interface, which looks like this:

public interface DefaultCacheManagerProvider {
  public CacheManager getDefaultCacheManager();
}

As this walk-through is based off of a CF app, we’ll create an implementation of this interface that parses VCAP_SERVICES. If you get adventurous and deploy your room elsewhere, you should be able to follow a similar pattern for retrieving the configuration of your endpoint from the environment.

So, to configure and create a CacheManager based on VCAP_SERVICES environment settings, we’ll do the following:

  1. Parse VCAP_SERVICES to obtain the host & credentials for Redis.

    Create an implementation of this interface that will parse VCAP_SERVICES, and configure a CacheManager for use by the annotations layer.

    Create a class in your room project that implements org.JSR-107.ri.annotations.DefaultCacheResolverFactory.DefaultCacheManagerProvider

    In the newly created class, add a private method parseVcapServices and have the implementation use JsonReader to read the JSON from the environment variable into a JsonObject, finally digging down through the JSON to get to the port,hostname and password fields stored within the rediscloud instance.

    The VCAP_SERVICES should look a little like:

    {
      "someotherservice": "[...]",
      "rediscloud": [
        {
          "name": "rediscloud-23",
          "label": "rediscloud",
          "plan": "30mb",
          "credentials": {
            "port": "6379",
            "hostname": "your.redis.server.hostname.com",
            "password": "your_redis_password"
          }
        }
      ]
    }
  2. Create the RedissonClient

    With the retrieved server details, you can create a ReddisonClient instance using code as follows:

    Config redissonConfig = new Config();
    redissonConfig.useSingleServer().setAddress(host+":"+port).setPassword(pwd);
    RedissonClient redisson = Redisson.create(redissonConfig);
  3. Create the CacheManager

    Finally you use the ReddisonClient, to create a CacheManager to satisfy the interface.

    CacheManager manager = new JCacheManager((Redisson)redisson,
                                             JCacheManager.class.getClassLoader(),
                                             null, null, null);

You are almost done, and the code would work as-is, but you need to be aware of a few issues.

  • Your implementation of DefaultCacheManagerProvider will be called each time a JSR-107 annotation is found.
  • Each time you do Redisson.create(…​)` you create an additional set of network connections to your Redis service instance
  • You only have a limited number of connections on the "free" tier of rediscloud.

So, if you plan to use more than a single annotated method, you will need to cache the RedissonClient and reuse it each time you are asked for a new CacheManager.

Here’s a full example implementation of a DefaultCacheManagerProvider that may be handy for you to reference. It parses VCAP_SERVICES and caches the RedissonClient instance as suggested.

Adding the META-INF/services entry

As mentioned earlier, the fork we are using of the JSR-107 CDI Module allows us to create the CacheManager for use by the annotations by supplying an implementation of a Java Service. We’ve created the implementation, and now we create the metadata that allows the implementation to be located at runtime.

Create a file in your Room project at src/main/webapp/META-INF/services and call it org.JSR-107.ri.annotations.DefaultCacheResolverFactory$DefaultCacheManagerProvider

Inside the file, place the full name for your DefaultCacheManagerProvider class, eg the example has the line saying…​

org.gameontext.sample.JSR-107defaultprovider.RedissonCacheManagerProvider

Congratulations! Your room is now able to use JSR-107 annotations, backed by your Redis service instance. Let’s look at a few ways we can use that in a room.

Secret Store

Using JSR-107 annotations, we will create a simple class that will allow players in the room to cache a "secret" that they can retrieve later.

The basic concept is simple; we’ll use a cache like a hashmap, and have it associate the players uniqueid, with the secret they will supply via a new Game On command /secret.

Creating the Store

The code for the secret store is deceptively simple;

@CacheDefaults(cacheName="secrets")
public class SecretDataBean {
    @CachePut
    public void setSecretForUser(@CacheKey String userid, @CacheValue String secret){
        //no-op
    }
    @CacheResult
    public String getSecretForUser(String userid){
        return null;
    }
}

The @CacheDefaults annotation sets up the class to use the cache called secrets. Using this annotation means we don’t need to specify the cache name on our other annotated methods.

The @CachePut annotated method will always update the cache. In this instance, we’re using the @CacheKey and @CacheValue annotations to have the cache values be identified straight from the method arguments themselves. Which means we don’t need a method body at all.

The @CacheResult annotation would normally be used to cache the result of invoking a method. It’s normal effect is to wrap the method invocation, and check the cache for a value with the key derived from the method arguments. If the cache has a value the method invocation is skipped entirely, otherwise the method is invoked, and the result of the method is set as the cached value, and returned to the caller.

In this example, we’re relying on the @CachePut to have updated the cache with the value we want to retrieve, so the only time the getSecretForUser method will actually execute is when there has been no value placed into the cache for the user via the put method. Effectively, this means the getSecretForUser method returns the "default" secret for when the user has not set one yet. Here we’re returning null which we’ll use in our command to identify there is no secret set for the user. But we could have chosen to do a database lookup, and retrieve a persisted key for the user.

Overall, this call conceptually acts a little bit like a Map, except the Map content is shared between all users of the Cache, which in this case could be multiple instances of our Room as it scales up under load. It can feel a bit strange to think of this as a Map, as it has no apparent storage within the class for the Keys & Values, because they are all managed by the Cache.

Adding a command to drive the Store

To test our Secret cache, lets add the new /secret command to our room to invoke it.

First, inject the SecretDataBean into the RoomImplementation class, add the annotated declaration near the top where other class variables are declared.

@Inject
protected SecretDataBean secret;

Then find the switch statement in the processCommand method, and add another case to the statement.

case "/secret":
    if (remainder == null) {
        String userSecret = secret.getSecretForUser(userId);
        if (userSecret == null) {
            endpoint.sendMessage(session,
                                 Message.createSpecificEvent(userId,
                                 "You apparently don't have a secret at the moment."+
                                 "Maybe you should set one with /secret ilikepie"));
        } else {
            endpoint.sendMessage(session,
                                 Message.createSpecificEvent(userId,
                                 "Your secret is currently '"+userSecret+"'"));
        }
    } else {
        secret.setSecretForUser(userId, remainder);
        endpoint.sendMessage(session,
                             Message.createSpecificEvent(userId,
                             "Your secret has been set to '"+remainder+"'"));
    }
    break;

Here when the command /secret is invoked with no arguments, we ask the secret store if it has a secret for the user, and output an appropriate message.

When invoked with arguments, we store that as the secret for the user.

Cache expiry

With our current Secret Store, we’ll hold onto the secret for the user until our Redis instance is restarted. This might not be quite what we want, if we had a large number of users who only try the Store once, we should clean up the Cache to remove old entries.

JSR-107 supports this concept by way of setting a CacheExpiry when the Cache is created. Unfortunately, when using the JSR-107 annotations, there is no handy 'expiry' annotation or attribute we can make use of. If we want to configure a cache used by the annotations, we are given a single option; the CacheResolverFactory.

A CacheResolverFactory can be set as an attribute for the various method annotations, and can also be set via the @CacheDefaults annotation. It has the responsibility of giving back a CacheResolver (which in turn gives back a Cache) for a given annotated method.

Here’s a simple CacheResolverFactory that will use the DefaultCacheManagerProvider we created earlier, to obtain a Redisson configured Cache with a 5 minute expiry. The Cache is then used to create a CacheResolver to return.

public class MyCacheResolverFactory implements CacheResolverFactory{

  CacheManager cacheManager = (new RedissonCacheManagerProvider())
                                 .getDefaultCacheManager();

  private Cache<?,?> getCache(String name){
    Cache<?, ?> cache = cacheManager.getCache(name);
    if (cache == null) {
      MutableConfiguration<Object, Object> config = getConfig();
      cacheManager.createCache(name, config);
      cache = cacheManager.getCache(name);
    }
  }

  private MutableConfiguration<Object,Object> getConfig(){
    MutableConfiguration<Object,Object> config = new MutableConfiguration<Object,Object>();
    config.setExpiryPolicyFactory(CreatedExpiryPolicy.factoryOf(Duration.FIVE_MINUTES));
    return config;
  }


  @Override
  public CacheResolver getCacheResolver(
         CacheMethodDetails<? extends Annotation> cacheMethodDetails) {

      Cache<?, ?> cache = getCache(cacheMethodDetails.getCacheName();)
      return new DefaultCacheResolver(cache);
  }

  @Override
  public CacheResolver getExceptionCacheResolver(
         CacheMethodDetails<CacheResult> cacheMethodDetails) {

      final CacheResult cacheResultAnnotation = cacheMethodDetails.getCacheAnnotation();
      Cache<?, ?> cache = getCache(cacheResultAnnotation.exceptionCacheName(););
      return new DefaultCacheResolver(cache);
  }
}

The code is pretty simple, the getCacheResolver and getExceptionCacheResolver methods obtain the cache name from the annotated method information, and then use the CacheManager from our DefaultCacheManagerProvider to lookup that cache. If the cache doesn’t exist, it’s created, and then it’s returned wrapped in a DefaultCacheResolver that will return the Cache when requested.

If we return to our SecretDataBean class and update it’s @CacheDefaults annotation to look like;

@CacheDefaults( cacheName="secrets" , cacheResolverFactory=MyCacheResolverFactory.class)

Then JSR-107 will now use our factory to obtain the cache used. Resulting in a 5 minute expiry time (from creation) for the Secrets in the Store.

To test it out, set a secret with the /secret command, then wait 6 minutes and ask for your secret.

Although we’ve used the cache here as a Secret Store, consider that the cache could be used to manage any sort of information we’d want to share between instances of our Service. You might use it to track Players in your room, or to assign virtual attributes to Players in your room, like health, or score. Or you might use it to track Room Inventory, or Inventory per Player. Or you might use it to manage state of items in your room, eg. If a light bulb in the room is on, or off.

Cache Based Lock

Because the Redis backed cache is common to each instance of the service using it, we can use it to implement a lock, so that only once instance of the service can manipulate some resource at the same time.

This would be especially handy for non atomic operations that span multiple remote cache states. Eg, transferring an object from Room Inventory to Player Inventory may involve removing the item from one cache and adding it to another. It’s important that the combined operation is performed by one instance, if two Players were to try to take the item at the same time, one should fail, rather than the object magically appearing in both Inventories.

@ApplicationScoped
@CacheDefaults( cacheName="locks" )
public class CacheBasedLockDataBean {
    //need to differentiate 'this jvm's locks from anyone-elses.
    private String uuid = UUID.randomUUID().toString();
    public String getUniqueId(){
        return uuid;
    }
    @CacheResult
    public String getReferenceLockForUserId(@CacheKey String item, String userid){
        //if the cache doesn't have an answer for this key, then it's not locked
        //at the mo, so we can return the requested user, which will be cached,
        //and returned if anyone else asks about it.
        return userid+getUniqueId();
    }
    @CacheRemove
    public void clearLockForRef(String item){
        //NO:OP, all the work done by the annotation.
    }
}

This creates a conceptual Map of "ItemId → (UserId + JVM_UUID)". If there is an entry for the ItemId, it means the item is considered locked by the UserID, with the lock held by the JVM with the corresponding UUID.

It works because if the ItemId is already locked by another player, or jvm, then the getReferenceLockForUserId method will return their userId+uuid. Only if the ItemId is currently not locked, will the method return a result indicating the lock was obtained successfully.

The lock release method clearLockForRef only has one task to do, and the @CacheRemove annotation takes care of it, removing the entry in the cache for the item id.

Obviously, this doesn’t make for a very intuitive API on our Lock, so you may wonder why we didn’t make these methods internal to the implementation, and expose a much nicer lock type API to callers. The answer is simple, the JSR-107 annotated methods must be public, only function if called from another Bean, not from within the same class.

To address the API issue, we’ll wrapper our Lock bean in another Bean that will offer a nicer interface to the other code.

@ApplicationScoped
public class CacheBasedLock {

    @Inject
    CacheBasedLockDataBean lockBean;

    /** Data store to track locks held by this JVM, in case we need to release them all */
    private Map<String,String> locksHeldByThisJVM = new ConcurrentHashMap<String,String>();

    /** Get lock for reference key, for requested userid */
    synchronized public boolean getLock(String reference, String userid){
        String currentLockedBy = lockBean.getReferenceLockForUserId(reference,userid);
        boolean success = currentLockedBy.equals(userid+lockBean.getUniqueId());
        if(success){
            locksHeldByThisJVM.put(reference, userid+lockBean.getUniqueId());
        }
        return success;
    }

    /** Release lock held by this JVM for reference key */
    synchronized public void releaseLock(String reference){
        lockBean.clearLockForRef(reference);
        locksHeldByThisJVM.remove(reference);
    }

    /** Utility method to release all locks we've acquired. */
    synchronized public void releaseAllLocksHeld(){
        for(String reference : locksHeldByThisJVM.keySet()){
            releaseLock(reference);
        }
    }
}

This simple wrapper injects itself with the Lock Bean, and offers a much simpler getLock method that can be used to attempt to acquire, or test if a lock is granted.

Additionally, it provides a little logic to allow us to clean up all locks held by the current instance of the app.

We can use our new Lock as follows;

@Inject
CacheBasedLock lock;

public testLock(String itemName, String userId){
  boolean gotLock = lock.getLock(itemName,userId);
  if(gotLock){
    try{
      //do something that needed lock.
    }finally{
      lock.releaseLock(itemName);
    }
  }

}

The Cache usage is totally hidden, but the effect is still present. Although this example doesn’t show how you can wait on the lock, it is possible to register CacheListeners that are invoked when the CacheContent changes, so you could add a Listener that would wait for a change signifying when the requested lock has been removed, and have it attempt to reacquire the lock.

We’ll show CacheListener usage over in the follow on JSR-107 API adventure =)

Working example repo.

For complete versions of the code discussed so far, check out my Sample JSR-107 Room. It does everything described here, and more, showing usage of both JSR-107 annotations, and direct API usage.

Suggested extensions

  • Implement room inventory / player inventory using a cache.
  • Implement item state using a cache.
  • Add a Game On command /lock to test the lock function.

Conclusion

Using Redis (via Redisson) as your JSR-107 implementation goes a long way to helping your service meet the 'stateless processes' goal for being a 12 factor app. Your app state, although feeling local, is actually managed by an instance of a stateful backing service (Redis).

JSR-107’s annotations help you to easily add caching type behavior to your service. Although they may seem a little restrictive at first, once you get to grips with them they quickly become a very powerful tool for managing information across multiple instances of a service. This approach is very effective for handling data that previously may have been stored within session storage.

Suggested further adventures.

You may want to take a look at the follow-on adventure "JSR-107 via API" which covers how to use JSR-107 without the annotations. (Keep an eye out for the "Redis via Redisson" adventure which will show a different spin on using Redis), or maybe the "Adding Items to a Room" adventure, that will give you additional ways to expose your Cache understanding within a Room.