Monday, July 13, 2015

Understanding Callable and Spring DeferredResult

1- Introduction

Asynchronous support introduced in Servlet 3.0 offers the possibility to process an HTTP request in another thread. This is specially interesting when you have a long running task, since while another thread processes this request, the container thread is freed and can continue serving other requests.

This topic has been explained many times, but there seems to be a little bit of confusion regarding those classes provided by the Spring framework which take advantage of this functionality. I am talking about returning Callable and DeferredResult from a @Controller.

In this post I will implement both examples in order to show its differences.

All the examples shown here consist on implementing a controller which will execute a long running task, and then return the result to the client. The long running task is processed by the TaskService:

The web application is built with Spring Boot. We will be executing the following class to run our examples:

The source code with all these examples can be found at the Github Spring-Rest repository.

2- Starting with a blocking controller

In this example, a request arrives to the controller. The servlet thread won't be released until the long running method is executed and we exit the @RequestMapping annotated method.

If we run this example at http://localhost:8080/block, looking at the logs, we can see that the servlet request is not released until the long running task has been processed (5 seconds later):

2015-07-12 12:41:11.849  [nio-8080-exec-6] x.s.web.controller.BlockingController    : Request received
2015-07-12 12:41:16.851  [nio-8080-exec-6] x.spring.web.service.TaskServiceImpl     : Slow task executed
2015-07-12 12:41:16.851  [nio-8080-exec-6] x.s.web.controller.BlockingController    : Servlet thread released

3- Returning Callable

In this example, instead of returning directly the result, we will return a Callable:

Returning Callable implies that Spring MVC will invoke the task defined in the Callable in a different thread. Spring will manage this thread by using a TaskExecutor. Before waiting for the long task to finish, the servlet thread will be released.

Let's take a look at the logs:

2015-07-12 13:07:07.012  [nio-8080-exec-5] x.s.w.c.AsyncCallableController          : Request received
2015-07-12 13:07:07.013  [nio-8080-exec-5] x.s.w.c.AsyncCallableController          : Servlet thread released
2015-07-12 13:07:12.014  [      MvcAsync2] x.spring.web.service.TaskServiceImpl     : Slow task executed

You can see that we have returned from the servlet before the long running task has finished executing. This doesn't mean the client has received a response. The communication with the client is still open waiting for the result, but the thread that received the request has been released and can serve another client's request.

4- Returning DeferredResult

First, we need to create a DeferredResult object. This object will be returned by the controller. What we will accomplish is the same with Callable, to release the servlet thread while we process the long running task in another thread.

So, what's the difference from Callable? The difference is this time the thread is managed by us. It is our responsibility to set the result of the DeferredResult in a different thread.

What we have done in this example, is to create an asynchronous task with CompletableFuture. This will create a new thread where our long running task will be executed. Is in this thread where we will set the result.

From which pool are we retrieving this new thread? By default, the supplyAsync method in CompletableFuture will run the task in the ForkJoin pool. If you want to use a different thread pool, you can pass an executor to the supplyAsync method:

If we run this example, we will get the same result as with Callable:

2015-07-12 13:28:08.433  [io-8080-exec-10] x.s.w.c.AsyncDeferredController          : Request received
2015-07-12 13:28:08.475  [io-8080-exec-10] x.s.w.c.AsyncDeferredController          : Servlet thread released
2015-07-12 13:28:13.469  [onPool-worker-1] x.spring.web.service.TaskServiceImpl     : Slow task executed

5- Conclusion

At a high level view, Callable and DeferredResult do the same exact thing, which is releasing the container thread and processing the long running task asynchronously in another thread. The difference is in who manages the thread executing the task.

I'm publishing my new posts on Google plus and Twitter. Follow me if you want to be updated with new content.

Tuesday, April 14, 2015

Configure a Spring JMS application with Spring Boot and annotation support

1   Introduction

In previous posts we learned how to configure a project using Spring JMS. If you check the article introduction to messaging with Spring JMS, you will notice that it is configured using XML. This article will take advantage of the improvements introduced in Spring 4.1 version, and configure a JMS project using Java config only.

In this example we will also see how easy it can be to configure the project by using Spring Boot.

Before we get started, just note that as usual, you can take a look at the source code of the project used in the examples below.

See the example project at github.

  1. Introduction.
  2. The example application.
  3. Setting up the project.
  4. A simple example with JMS listener.
  5. Sending a response to another queue with @SendTo.
  6. Conclusion.

2   The example application

The application uses a Client service to send orders to a JMS queue, where a JMS listener will be registered and handle these orders. Once received, the listener will store the order through the Store service:

We will use the Order class to create orders:

Before moving on to the first example, we will first explore how the project structure is built.

3   Setting up the project

3.1   Configuring pom.xml

The first thing to do is to define the artifact spring-boot-starter-parent as our parent pom.

This parent basically sets several Maven defaults and provides the dependency management for the main dependencies that we will use, like the Spring version (which is 4.1.6).

It is important to note that this parent pom defines the version of many libraries but it does not add any dependency to our project. So don’t worry about getting libraries you won’t use.

The next step is to set the basic dependencies for Spring Boot:

In addition to the core Spring libraries, this dependency will bring the auto configuration functionality of Spring Boot. This will allow the framework to try to automatically set up the configuration based on the dependencies you add.

Finally, we will add the Spring JMS dependency and the ActiveMQ message broker, leaving the whole pom.xml as follows:

3.2   Spring Configuration with Java Config

We used @SpringBootApplication instead of the usual @Configuration annotation. This Spring Boot annotation is also annotated with @Configuration. In addition, it sets other configuration like Spring Boot auto configuration:

The configuration class does not need to define any bean. All the configuration is automatically set by Spring Boot. Regarding the connection factory, Spring Boot will detect that I included the ActiveMQ dependency on the classpath and will start and configure an embedded broker.

If you need to specify a different broker url, you can declare it in the properties. Check ActiveMQ support section for further detail.

It is all set now. We will see how to configure a JMS listener in the example in the next section, since it is configured with an annotation.

4   A simple example with JMS listener

4.1   Sending an order to a JMS queue

The ClientService class is responsible for sending a new order to the JMS queue. In order to accomplish this, it uses a JmsTemplate:

Here, we use a JmsTemplate to convert our Order instance and send it to the JMS queue. If you prefer to directly send a message through the send message, you can instead use the new JmsMessagingTemplate. This is preferable since it uses the more standardized Message class.

4.2   Receiving an order sent to the JMS queue

Registering a JMS listener to a JMS listener container is as simple as adding the @JmsListener annotation to the method we want to use. This will create a JMS listener container under the covers that will receive messages sent to the specified queue and delegate them to our listener class:

The StoreService receives the order and saves it to a list of received orders:

4.3   Testing the application

Now let’s add a test to check if we did everything correctly:

5   Sending a response to another queue with @SendTo

Another addition to Spring JMS is the @SendTo annotation. This annotation allows a listener to send a message to another queue. For example, the following listener receives an order from the “in.queue” and after storing the order, sends a confirmation to the “out.queue”.

There, we have another listener registered that will process this confirmation id:

6   Conclusion

With annotation support, it is now much easier to configure a Spring JMS application, taking advantage of asynchronous message retrieval using annotated JMS listeners.

I'm publishing my new posts on Google plus and Twitter. Follow me if you want to be updated with new content.

Sunday, March 8, 2015

Improving performance: non-blocking processing of streams

1   Introduction

Imagine we have an application that needs to access an external web service in order to gather information about clients and then process it. More specifically, we can’t get all this information in a single invocation. If we want to look up different clients, we will need several invocations.

As shown in the graphic below, the example application will retrieve information about several clients, group them in a list and then process it to calculate the total amount of its purchases:

In this post, we will see different ways of gathering the information and which one is the best in terms of performance.

This is a Java related post. However, we will use the Spring framework to invoke a RESTful web service.

  1. Introduction
  2. Explaining the example
  3. First attempt: Sequential stream
  4. Improving performance: Parallel stream
  5. Non-blocking processing with CompletableFuture
  6. Conclusion

The source code can be found at the Java 8 GitHub repository.

Additionally, you can access the source code of the web application exposing the RESTful web service at this repository.

2   Explaining the example

In our application, we have a list of 20 ids representing clients we want to retrieve from a web service. After retrieving all the clients, we will look up at what did every client purchase and sum them up to compute what is the total amount of money spent by all the clients.

There is one problem though, this web service only allows to retrieve one client at each invocation, so we will need to invoke the service twenty times. In addition, the web service is a little bit slow, taking at least two seconds to respond to a request.

If we take a look at the application implementing the web service, we can see that invocations are handled by the ClientController class:

A Thread.sleep is used to simulate the slowness in responding.

The domain class (Client) contains the information we need; how much money has a client spent:

3   First attempt: Sequential stream

In this first example we will sequentially invoke the service to get the information of all twenty clients:

Sequential | Total time: 42284 ms
Total purchases: 20.0

The execution of this program takes 42 seconds approximately. This is too much time. Let’s see if we can improve its performance.

4   Improving performance: Parallel stream

Java 8 allows us to split a stream into chunks and process each one in a separate thread. What we need to do is simply create the stream in the previous example as a parallel stream.

You should take into account that each chunk will be executed in its thread asynchronously, so the order in which the chunks are processed must not matter. In our case, we are summing the purchases, so we can do it.

Let’s try this:

Parallel | Total time: 6336 ms
Total purchases: 20.0

Wow, that’s a big improvement! But what does this number come from?

Parallel streams internally use the ForkJoinPool, which is the pool used by the ForkJoin framework introduced in Java 7. By default, the pool uses as many threads as your machine's processors can handle. My laptop is a quad core that can handle 8 threads (you can check this by invoking Runtime.getRuntime.availableProcessors), so it can make 8 invocations to the web service in parallel. Since we need 20 invocations, it will need at least 3 "rounds":

Ok, so from 40 seconds to 6 is quite a good improvement but, can we still improve it further? The answer is yes.

5   Non-blocking processing with CompletableFuture

Let’s analise the previous solution.

We send 8 threads invoking each one the web service, but while the service is processing the request (two whole seconds), our processors are doing nothing but waiting (this is a IO operation). Until these requests don’t come back, we won’t be able to send more requests.

The question is, what if we could send all 20 requests asynchronously, freeing our processors and process each response when is available? This is where CompletableFuture comes to the rescue:

Async with executor | Total time: 2192 ms
Total purchases: 20.0

It took a third of the time spent in the previous example.

We sent all 20 requests at the same time, so the time spent in IO operations is spent only once. As soon as responses come by, we process them quickly.

It is important the use of the executor service, set as an optional second parameter of the supplyAsync method. We specified a pool of a hundred threads so we could send 100 requests at the same time. If we don’t specify an executor, the ForkJoin pool will be used by default.

You can try to remove the executor and you will see the same performance as in the parallel example.

6   Conclusion

We have seen that when executing operations that do not involve computing (like IO operations) we can use the CompletableFuture class to take advantage of our processors and improve the performance of our applications.

I'm publishing my new posts on Google plus and Twitter. Follow me if you want to be updated with new content.

Sunday, February 15, 2015

Java Concurrency Tutorial - Locking: Explicit locks

1   Introduction

In many cases, using implicit locking is enough. Other times, we will need more complex functionalities. In such cases, java.util.concurrent.locks package provides us with lock objects. When it comes to memory synchronization, the internal mechanism of these locks is the same as with implicit locks. The difference is that explicit locks offer additional features.

The main advantages or improvements over implicit synchronization are:

  • Separation of locks by read or write.
  • Some locks allow concurrent access to a shared resource (ReadWriteLock).
  • Different ways of acquiring a lock:
    • Blocking: lock()
    • Non-blocking: tryLock()
    • Interruptible: lockInterruptibly()

2   Classification of lock objects

Lock objects implement one of the following two interfaces:

  • Lock: Defines the basic functionalities that a lock object must implement. Basically, this means acquiring and releasing the lock. In contrast to implicit locks, this one allows the acquisition of a lock in a non-blocking or interruptible way (additionally to the blocking way). Main implementations:
    • ReentrantLock
    • ReadLock (used by ReentrantReadWriteLock)
    • WriteLock (used by ReentrantReadWriteLock)

  • ReadWriteLock: It keeps a pair of locks, one for read-only operations and another one for writing. The read lock can be acquired simultaneously by different reader threads (as long as the resource isn’t already acquired by a write lock), while the write lock is exclusive. In this way, we can have several threads reading the resource concurrently as long as there is not a writing operation. Main implementations:
    • ReentrantReadWriteLock

The following class diagram shows the relation among the different lock classes:

3   ReentrantLock

This lock works the same way as the synchronized block; one thread acquires the lock as long as it is not already acquired by another thread, and it does not release it until unlock is invoked. If the lock is already acquired by another thread, then the thread trying to acquire it becomes blocked until the other thread releases it.

We are going to start with a simple example without locking, and then we will add a reentrant lock to see how it works.

Since the code above is not synchronized, threads will be interleaved. Let’s see the output:

Thread-2 - 1
Thread-1 - 1
Thread-1 - 2
Thread-1 - 3
Thread-2 - 2
Thread-2 - 3

Now, we will add a reentrant lock in order to serialize the access to the run method:

The above code will safely be executed without threads being interleaved. You may realize that we could have used a synchronized block and the effect would be the same. The question that arises now is what advantages does the reentrant lock provides us?

The main advantages of using this type of lock are described below:

  • Additional ways of acquiring the lock are provided by implementing Lock interface:
    • lockInterruptibly: The current thread will try to acquire de lock and become blocked if another thread owns the lock, like with the lock() method. However, if another thread interrupts the current thread, the acquisition will be cancelled.
    • tryLock: It will try to acquire the lock and return immediately, regardless of the lock status. This will prevent the current thread from being blocked if the lock is already acquired by another thread. You can also set the time the current thread will wait before returning (we will see an example of this).
    • newCondition: Allows the thread which owns the lock to wait for a specified condition.

  • Additional methods provided by the ReentrantLock class, primarily for monitoring or testing. For example, getHoldCount or isHeldByCurrentThread methods.

Let’s look at an example using tryLock before moving on to the next lock class.

3.1   Trying lock acquisition

In the following example, we have got two threads, trying to acquire the same two locks.

One thread acquires lock2 and then it blocks trying to acquire lock1:

Another thread, acquires lock1 and then it tries to acquire lock2.

Using the standard lock method, this would cause a dead lock, since each thread would be waiting forever for the other to release the lock. However, this time we are trying to acquire it with tryLock specifying a timeout. If it doesn’t succeed after four seconds, it will cancel the action and release the first lock. This will allow the other thread to unblock and acquire both locks.

Let’s see the full example:

If we execute the code it will result in the following output:

13:06:38,654|Thread-2|Trying to acquire lock2...
13:06:38,654|Thread-1|Trying to acquire lock1...
13:06:38,655|Thread-2|Lock2 acquired. Trying to acquire lock1...
13:06:38,655|Thread-1|Lock1 acquired. Trying to acquire lock2...
13:06:42,658|Thread-1|Failed acquiring lock2. Releasing lock1
13:06:42,658|Thread-2|Both locks acquired

After the fourth line, each thread has acquired one lock and is blocked trying to acquire the other lock. At the next line, you can notice the four second lapse. Since we reached the timeout, the first thread fails to acquire the lock and releases the one it had already acquired, allowing the second thread to continue.

4   ReentrantReadWriteLock

This type of lock keeps a pair of internal locks (a ReadLock and a WriteLock). As explained with the interface, this lock allows several threads to read from the resource concurrently. This is specially convenient when having  a resource that has frequent reads but few writes. As long as there isn’t a thread that needs to write, the resource will be concurrently accessed.

The following example shows three threads concurrently reading from a shared resource. When a fourth thread needs to write, it will exclusively lock the resource, preventing reading threads from accessing it while it is writing. Once the write finishes and the lock is released, all reader threads will continue to access the resource concurrently:

The console output shows the result:

11:55:01,632|pool-1-thread-1|Read lock acquired
11:55:01,632|pool-1-thread-2|Read lock acquired
11:55:01,632|pool-1-thread-3|Read lock acquired
11:55:04,633|pool-1-thread-3|Reading data: default value
11:55:04,633|pool-1-thread-1|Reading data: default value
11:55:04,633|pool-1-thread-2|Reading data: default value
11:55:04,634|pool-1-thread-4|Write lock acquired
11:55:07,634|pool-1-thread-4|Writing data: changed value
11:55:07,634|pool-1-thread-3|Read lock acquired
11:55:07,635|pool-1-thread-1|Read lock acquired
11:55:07,635|pool-1-thread-2|Read lock acquired
11:55:10,636|pool-1-thread-3|Reading data: changed value
11:55:10,636|pool-1-thread-1|Reading data: changed value
11:55:10,636|pool-1-thread-2|Reading data: changed value

As you can see, when writer thread acquires the write lock (thread-4), no other threads can access the resource.

5   Conclusion

This post shows which are the main implementations of explicit locks and explains some of its improved features with respect to implicit locking.

This post is part of the Java Concurrency Tutorial series. Check here to read the rest of the tutorial.

You can find the source code at Github.

I'm publishing my new posts on Google plus and Twitter. Follow me if you want to be updated with new content.