JDK 7 introduced a interesting concurrency concept of mix between CountDownLatch and CyclicBarrier mechanisms. It's name - Phaser.
Data Engineering Design Patterns
Looking for a book that defines and solves most common data engineering problems? I wrote
one on that topic! You can read it online
on the O'Reilly platform,
or get a print copy on Amazon.
I also help solve your data engineering problems 👉 contact@waitingforcode.com 📩
This article will describe this class more in details. At the begin, we'll focus on Phaser specificity, as available methods and differences between it and CountDownLatch and CyclicBarrier classes. The second part will show an example of the code using Phaser.
What is Phaser ?
According to Phaser's Javadoc, Phaser is a class which works similarly to CyclicBarrier and CountDownLatch. As these two last classes, Phaser also allows several threads to wait until the condition is accomplished before continue. A condition can be, for example, that all "phased" threads terminate theirs tasks. In exactly the same way works CountDownLatch which plays with await() and countDown() methods to achieve the same effect.
So, which are the differences between Phaser and CyclicBarrier and CountDownLatch ? Let's enumerate them:
- Phaser can handle several threads dynamically. Ie. the number of threads participating in Phaser job can change every time, event after the initialization of Phaser object. It's not the case for CountDownLatch and CyclichBarrier which are constructed with fixed number of participants without the possibility to change it.
- Unlike CountDownLatch, Phaser can be reused. We cna use the same Phaser instance in whole application.
- Phaser's threads can continue to execute theirs tasks after signaling theirs arrivals (arrive() method). It's not the case of CyclicBarrier.
If these concepts seem clear, we can explore available methods of Phaser:
- Arrival methods
We distinguish 3 of arrival methods. The simplest one, arrive(), signals that one task reached the barrier. If arrive() invocation is not the last method call of this task, it will continue to execute. The same behavior occurs on arriveAndDeregister(). Here in additionally, task is unregistered from Phaser and continues to execute the code after reaching the Phaser. It's not the case of arriveAndAwaitAdvance() where the task will wait that all tasks reach the Phaser before continue.
- Waiting methods
Phaser's tasks can wait for another tasks more explicitly than with methods saw in the previous point. For example, with awaitAdvance(int phase) method, task waits for the phase passed in parameter is terminated. A very similar method is awaitAdvanceInterruptibly(int phase). But unlike the previous one, it throws an InterruptedException when the thread is interrupted while waiting.
Even more complex waiting method is awaitAdvanceInterruptibly(int phase, long timeout, TimeUnit unit). It allows to wait while the time specified in two last parameters. If the timeout occurs, an TimeoutException is thrown. If the thread is interrupted, InterruptedException is thrown.
- Termination methods
To see if Phaser is terminated, we can simply call isTerminated(). To force it to terminate, you should callforceTermination().
Registering
As we mentioned at the begin, Phaser allows dynamic registering. To register new task to Phaser, we can invoke register() or bulkRegister(int parties) method. The first one adds only 1 task while in the second we can pass any superior number than 0 (otherwise, IllegalArgumentException is thrown).
Example of use of Phaser
In our example, we'll inspect Phaser behavior in several angles. The first one will consist to show how we can execute customized parts of code according to just reached phase. You'll note it with awaitAdvanceInterruptibly calls. The second approached aspect is the reusability. In our testBasics case we use one Phaser object to make two grouped tasks. Each task is executed by 3 workers. In additionally, we add one phaser which will block the execution of main thread thanks to arriveAndAwaitAdvance method. And finally, we'll see that the reusability is possible with arriveAndDeregister method which decrements the counter of registered tasks. Otherwise, for example with simple arrive() method, the counter won't decrement and the program risks to run infinitely.
This is the test case:
public class PhaserTest {
@Test
public void testBasics() {
final Phaser phaser = new Phaser();
/**
* Current phase is 0 because they're no arrive invocations before.
*/
assertTrue("Current phase should be 0 but is "+phaser.getPhase(), phaser.getPhase() == 0);
/**
* This block will try a TimeoutException. Why ? Because the phase 0 isn't advanced.
* It will be advanced when all registered tasks will invoke arrive* method.
*/
boolean wasTe = false;
try {
phaser.awaitAdvanceInterruptibly(0, 2, TimeUnit.SECONDS);
} catch (TimeoutException te) {
wasTe = true;
}
assertTrue("TimeoutException should be thrown but it wasn't", wasTe);
Thread worker1 = new Thread(new Worker(phaser, "#1"));
Thread worker2 = new Thread(new Worker(phaser, "#2"));
Thread worker3 = new Thread(new Worker(phaser, "#3"));
worker1.start();
final Map<String, String> crashContainer = new HashMap<String, String>();
/**
* Note that all phaser tasks call arrive* after 2 seconds. Here, we're
* waiting 3 seconds before to check if 0 phase was reached. This code
* shouldn't fail because phase 0 should be done before this timeout.
*/
new Thread(new Runnable() {
@Override
public void run() {
try {
phaser.awaitAdvanceInterruptibly(0, 3, TimeUnit.SECONDS);
crashContainer.put("testOk3Sec", "OK");
} catch (Exception e) {
crashContainer.put("testOk3Sec", "NOK");
}
}
}).start();
worker2.start();
worker3.start();
phaser.arriveAndAwaitAdvance();
/**
* Because phaser can be reused, we can run once again our workers
* with the same Phaser instance. However, be careful about the number
* of registered tasks. If you register new tasks without unregistering
* the old ones (already terminated), your code will never stop. It's
* the reason why in this sample we invoke awaitAndDeregister in
* tasks run() method.
*/
Thread worker4 = new Thread(new Worker(phaser, "#4"));
Thread worker5 = new Thread(new Worker(phaser, "#5"));
Thread worker6 = new Thread(new Worker(phaser, "#6"));
worker4.start();
worker5.start();
worker6.start();
/**
* We test again the execution of specific code after the terminate of the 1st phase.
*/
new Thread(new Runnable() {
@Override
public void run() {
try {
phaser.awaitAdvanceInterruptibly(1, 3, TimeUnit.SECONDS);
crashContainer.put("testOk3SecP1", "OK");
} catch (Exception e) {
crashContainer.put("testOk3SecP1", "NOK");
}
}
}).start();
phaser.arriveAndAwaitAdvance();
// Sleep by security
TimeUnit.SECONDS.sleep(2);
assertTrue("Phase 0: 3 seconds timeout method shouldn't fail but it was",
crashContainer.get("testOk3Sec").equals("OK"));
assertTrue("Phase 1: 3 seconds timeout method shouldn't fail but it was",
crashContainer.get("testOk3SecP1").equals("OK"));
}
}
class Worker implements Runnable {
private Phaser phaser;
private String name;
public Worker(Phaser phaser, String name) {
this.phaser = phaser;
this.name = name;
phaser.register();
}
@Override
public void run() {
try {
Thread.sleep(2000);
} catch (InterruptedException e) {
e.printStackTrace();
}
this.phaser.arriveAndDeregister();
}
}
Phaser shows that is a great mix of CyclicBarrier and CountDownLatch objects. It's a reusable barrier which can change the number of participating threads dynamically. Thanks to it, we can also execute some specific tasks to realized steps thanks to awaitAdvanceInterruptibly method.
Consulting
With nearly 16 years of experience, including 8 as data engineer, I offer expert consulting to design and optimize scalable data solutions.
As an O’Reilly author, Data+AI Summit speaker, and blogger, I bring cutting-edge insights to modernize infrastructure, build robust pipelines, and
drive data-driven decision-making. Let's transform your data challenges into opportunities—reach out to elevate your data engineering game today!
👉 contact@waitingforcode.com
đź”— past projects
