Playing with Hazelcast on Payara

I have been reading a bit about Hazelcast, came up to chapter 5 in http://hazelcast.org/mastering-hazelcast/, but as I am a bit impatient and wanted to try out some stuff before reading the whole book and thinking about possibilities to use Hazelcast in my current work, this Weekend I set up to solve the following:

• Use Hazelcast IMap in a Java EE container
• Use MapLoader to load into a IMap with JPA
• As far as possible, use CDI

Payara comes with a Hazelcast, but solving this tasks may not be so straightforward, and I will try to explain why.

As usual I use a simple blog Post RESTful service as an example.

#%RAML 0.8 title: Blog post service baseUri: http://localhost:8080/hazelblog/api version: v1 /posts: get: description: Gets a list of "Posts" queryParameters: author: displayName: Author type: string description: Name of the author example: @ironic responses: 200: body: application/json: example: | [ {"author":"@ironic","content":"Nothing interesting","id":1,"title":"First post"}, {"author":"@ironic","content":"Again nothing interesting","id":2,"title":"Second post"} ] post: description: Creates a post body: application/json: example: | {"author":"@ironic","content":"Nothing interesting", "title":"Test post"} /{id}: get: description: View of a post responses: 200: body: application/json: example: | {"author":"@ironic","content":"Nothing interesting","id":1,"title":"First post"} delete: description: Deletes a post

Using Hazlecast in Payara

Even if Payara comes with an embedded Hazelcast, I will not use it by letting Payara start it, and the reason is what I found when playing around with EntryListener. I will however use the Hazelcast that is provided, but will not activate it as written here: https://github.com/payara/Payara/wiki/Hazelcast-%28Payara-4.1.153%29 but more with ideas as written here: http://blog.modernjava.info/2014/10/using-hazelcast-as-jcache-provider-with.html.

To start with, this is what we will need in the pom.xml,

<dependencies> <dependency> <groupid>javax</groupId> <artifactid>javaee-web-api</artifactId> <version>7.0</version> <scope>provided</scope> </dependency> <dependency> <groupid>com.hazelcast</groupId> <artifactid>hazelcast</artifactId> <version>3.5</version> <scope>provided</scope> </dependency> </dependencies>

I am using Hazelcast 3.5 since that is the one that comes with Payara 4.153.

To fire up Hazelcast we will use an annotation and a provider so we can use CDI where we want to use stuff from Hazelcast.

package se.ironic.hazelblog; import static java.lang.annotation.ElementType.*; import java.lang.annotation.Retention; import static java.lang.annotation.RetentionPolicy.RUNTIME; import java.lang.annotation.Target; import javax.inject.Qualifier; @Qualifier @Retention(RUNTIME) @Target({TYPE, METHOD, FIELD, PARAMETER}) public @interface Hazelcast { }
package se.ironic.hazelblog; import com.hazelcast.config.Config; import com.hazelcast.config.MapStoreConfig; import com.hazelcast.core.HazelcastInstance; import javax.enterprise.context.ApplicationScoped; import javax.enterprise.inject.Disposes; import javax.enterprise.inject.Produces; import javax.inject.Singleton; @ApplicationScoped public class HazelcastProvider { @Produces @Singleton @Hazelcast public HazelcastInstance create() { System.out.println("Creating Hazelcast Instance"); ClassLoader classLoader = Thread.currentThread().getContextClassLoader(); Config config = new Config(); MapStoreConfig storeConfig = new MapStoreConfig(); storeConfig.setEnabled(true); storeConfig.setClassName("se.ironic.hazelblog.PostLoader"); config.getMapConfig("posts").setMapStoreConfig(storeConfig); HazelcastInstance newInst = com.hazelcast.core.Hazelcast.newHazelcastInstance(config); return newInst; } public void close(@Disposes @Hazelcast HazelcastInstance instance) { System.out.println("Shutting down Hazelcast...."); instance.shutdown(); } }

The reason for doing like this is that I noticed by playing around with EntryListener that by letting Payara firing up Hazelcast I cannot really add listeners, since Hazelcast then know nothing about my classes (resulting in ClassNotFoundException), so to resolve this we need to provide the class loader from our web-application to Hazelcast. To do this we need to start it "manually" and provide the class loader so Hazelcast knows about them. You can also see we configure Hazelcast with a MapLoader, more on that later.

REST-resource

I will not share the full code since you can find it on Bitbucket. In PostResource we use CDI to get hold of the Hazelcast, it will initialise on first use, so if no pre-warming occurs (like using a startup bean) it will take 5-10 seconds.

@Inject @Hazelcast private HazelcastInstance hazelcastInstance;
And use it
@GET @Produces(MediaType.APPLICATION_JSON) @Path("{id}") public Post get(@PathParam("id") Long id) { IMap<Long, Post> map = hazelcastInstance.getMap("posts"); Post post = map.get(id); if(post == null) { throw new WebApplicationException("No post found!", Response.Status.NOT_FOUND); } return post; }
For the creation of a Post we can use a trick to load it in the map by trying to get it after the save. That is we don't put it in the map directly and use a MapStore to save it (we don't know the id until after the persist). But since this is about using MapLoader, we will use that and not put it (as indicated in the comments)
@POST @Consumes(MediaType.APPLICATION_JSON) public void post(Post post) { postRepository.persist(post); // need to "touch" it to load it to the map. // could also have use put since we have id after persist. IMap<Long, Post> map = hazelcastInstance.getMap("posts"); map.get(post.getId()); }

MapLoader using JPA

Actually, I am not using it directly from the MapLoader, but injecting a @Statless bean from it, that does the actual JPA stuff.
As the MapLoader lives in Hazelcast and is created by it, we cannot use @Inject or even make it a managed bean (well we can, but it will not help since it is not using the jave ee container to handle it). But since CDI 1.1 we can inject using the CDI-class.

This is how the map loader I create looks like:
package se.ironic.hazelblog; import com.hazelcast.core.MapLoader; import java.util.Collection; import java.util.Collections; import java.util.HashMap; import java.util.List; import java.util.Map; import javax.enterprise.inject.spi.CDI; public class PostLoader implements MapLoader<Long, Post> { @Override public Post load(Long id) { PostRepository repo = CDI.current().select(PostRepository.class).get(); Post post = repo.find(id); return post; } @Override public Map<Long, Post> loadAll(Collection<long> ids) { Map<Long, Post> all = new HashMap<>(); PostRepository repo = CDI.current().select(PostRepository.class).get(); Map<String, Object> params = new HashMap<>(); params.put("ids", ids); List<post> posts = repo.findByQuery("SELECT p FROM Post p WHERE p.id in :ids", params); for (Post post : posts) { all.put(post.getId(), post); System.out.println("Post: " + post); } return all; } @Override public Iterable<long> loadAllKeys() { PostRepository repo = CDI.current().select(PostRepository.class).get(); List<long> ids = repo.findByQuery("SELECT p.id FROM Post p", Collections.EMPTY_MAP); return ids; } }
Notice the code to inject the PostRepository (which is the bean using JPA with an EntityManager). This will work since Hazelcast is loaded from the container with our classes.
PostRepository repo = CDI.current().select(PostRepository.class).get();

Some notes

I have implemented loadAllKeys() and loadAll(...), this is probably fine since I do not have so many posts, but for a huge database and if you do not want them all in memory, you may leave them empty and return null for keys or you will end up waiting for a while. I played with a criteria search on the map

@GET @Produces(MediaType.APPLICATION_JSON) public Collection<post> list() { String author = uriInfo.getQueryParameters().getFirst("author"); if(author != null) { System.out.println("Searching on map... author:" + author); IMap<Long, Post> map = hazelcastInstance.getMap("posts"); EntryObject entryObject = new PredicateBuilder().getEntryObject(); PredicateBuilder predicate = entryObject.get("author").equal(author); Collection<post> values = map.values(predicate); return values; } return Collections.EMPTY_LIST; }
and found out that by doing that, Hazelcast will try to load all keys from the database before it searcher memory.
However, this is done only once, if you add items to the database, it will not get loaded if you do not do anything to tell the map, as you have seen in the "@POST" or put it there yourself, subsequent searches on the map will not try to load items not there from the first loadAll.

Payara does not really matter in the context, other than having a Hazelcast and being a Java EE container. But I also use it since it have a default datasource for Derby, so I kind of used that to avoid setting up a database, and load a few posts to get started.If you use another container, you may have to change the hazelnuts dependency to be "compile" instead of "provided" if that container donate have it. You may also need to set up a datasource.


The code can be found here: [https://bitbucket.org/pejomstd/hazelblog]