A not thread safe counter

As an example we use the following jersey rest service. It consists of a resource which increments a counter for each post call and returns the new value:

@Path("counter")
public class Counter {
     private static int i = 0;
    @POST
    @Produces(MediaType.TEXT_PLAIN)
    public String addOne() {
    	return new Integer(i++).toString();
    }
}

This is clearly not thread safe. The access to the variable counter is not synchronized, which will lead to a race condition, if the method "addOne" is called from too many threads in parallel. Let us see, if we can detect this bug with a test.

How to test

To test if this service is thread safe, we need a multi threaded test, like the following:

@RunWith(ConcurrentTestRunner.class)
public class CounterTest {
    private HttpServer server;
    private WebTarget target;
    @Before
    public void setUp() throws Exception {
        server = Main.startServer();
        Client c = ClientBuilder.newClient();
        target = c.target(Main.BASE_URI);
    }
    @After
    public void tearDown() throws Exception {
        server.shutdown();
    }
    @Test
    public void testAddOne() {	
    		String responseMsg = target.path("counter").request().post(Entity.json(null) , String.class);
    		/*
    		 * 
    		 * Checking the responseMsg left out for brevity...
    		 * 
    		 */  
    }
}

The concurrent test runner runs the test method in parallel with THREAD_COUNT threads. To detect race conditions, we need a tool, which can detect race conditions during tests. One such tool is vmlens. We can enable it, by adding the vmlens agent path to the vm arguments. After running we will see the race condition in vmlens:

After we have found the race condition, we want to fix the race.

Making the rest service thread safe

The easiest way to do this, is to use a java.util.concurrent.atomic.AtomicInteger. AtomicInteger uses a volatile Field internally, making updates visible to all threads. And the used method "addAndGet" is made atomic, by using compareAndSet.

@Path("counter")
public class Counter {
     private static AtomicInteger i = new AtomicInteger();
    @POST
    @Produces(MediaType.TEXT_PLAIN)
    public String addOne() {
    	return new Integer(i.incrementAndGet()).toString();
    }
}

Conclusion

To test a multi threaded java rest service we need two things, a multi threaded test and a tool which can detect java race conditions. For the multi threaded test I used concurrent-junit, for the race condition detection I used vmlens. If you have a question or remark please add a comment below.