Threads communication happens primarily by sharing access to fields and objects. Although extremely efficient, this form of communication is prone to errors such as thread interference and memory consistency. Synchronization is a tool that helps to prevent such errors.
However, synchronization does not come for free and can introduce latency when accessing a lock or object that is currently being held by another thread. The waiting thread cannot use that object until the other thread releases the lock on the object. This condition is known as thread contention. It may also lead to deadlocks and livelocks.
In this post, we will explore the different options that Java provides to deal with threads synchronization.
Java provides a series of mechanisms to handle basic thread synchronization and coordination. It supports fine-grained object access synchronization through synchronized methods and synchronized statements. Basic thread coordination can be accomplished through guarded blocks. All mentioned mechanisms are built around acquiring and releasing the object's intrinsic lock.
Every Java object has an associated intrinsic lock. A thread that needs exclusive access to an object's fields has to acquire the object's lock before accessing them, and then releasing the intrinsic lock once it is done. Other threads trying to access the object will block until the thread holding the lock releases it.
When a thread invokes a synchronized method, it acquires the intrinsic lock for that method's object and releases it when the methods returns. The lock is released even if the method returns due to an uncaught exception. If done in a static method, the thread acquires the lock for the class object associated with the class.
Provides a more fine-grained synchronization mechanism. Synchronized statements must specify the object that provides the intrinsic lock. Synchronizing over separated lock objects can provide fields synchronization, without forcing synchronization between methods calls.
As mentioned earlier, guarded blocks provide support for thread coordination. Guarded blocks are part of every Java object, and can be constructed using the wait, notify and notifyAll methods.
The wait method suspend the current thread. When a thread invokes wait, it must own the object's intrinsic lock, that is why calls to wait are usually wrapped in a synchronized method or statement. The invocation of the wait method suspends the thread execution and releases the lock.
At some point, another thread will acquire the object's intrinsic lock and invoke notifyAll to inform all threads waiting that something important has happened. After the second thread has released the lock, the waiting threads will reacquire the lock and resume execution by returning from the wait invocation.
Notify wakes up a single thread. The concrete thread that is woken up can not be specified, therefore, it is useful only if we do not care which thread is woken up.
The Java Synchronizers
Java also provide five classes for common special-purpose synchronization.
The CountDownLatch class allows one or more threads to wait until a set of operations in other threads complete. It is initialised with a count number.
The await method blocks until the count reaches zero.
The countDown method decrements the count.
When the await method returns all waiting threads are released and subsequent invocations to await return immediately. The count cannot be reset.
The Semaphore is used to restrict thread access to a certain resource. It is initialised with a number of permits.
The acquire method blocks until a permit is available and takes it.
The release method adds a permit, releasing a blocking acquirer.
Note that calls to release does not have to be made by the same thread that called acquire. A semaphore can be fair or unfair. If fair, then the threads acquire permits in a FIFO fashion.
Although at first it may seem similar to the CountDownLatch its purpose is completely different.
The CyclicBarrier is built around the concept of parties. It allows threads to wait for each other to reach a common barrier point.
The await method blocks until all parties arrive. It behaves somehow as the inverse of the CountDownLatch. After N awaits it continues.
It has support for an optional runnable that runs once per barrier point. After the last party arrives, but before they are released. It is usually used to update shared-state between threads. It is cyclic because it can be reused after threads are released.
The Exchanger is a synchronization point at which two threads can exchange information.
Threads block until its counterpart presents its information. The same behaviour occurs at both sides.
The Phaser is a reusable barrier, similar to CountDownLatch and CyclirBarrier, but much more flexible.
In phaser, the number of registered parties is not fixed at creation time. Parties can register at any time through register or bulkRegister methods. Parties can deregister upon arrival with arriveAndDeregister.
It offers several methods for synchronization. The arriveAndAwaitAdvance method behaves the same way as CycleBarrierawait method does. arrive and arriveAndDeregister record arrival, but don't block. awaitAdvance blocks until all parties arrive.
It can be terminated, forcing all synchronization methods to return. Can be forced through the forceTermination method.
It also provides support for monitoring its state. It is noteworthy mentioning that synchronization methods can be called only by registered parties, whilst state can be monitored by any caller. Monitoring methods include getRegisteredParties and getArrivedParties among others.
Multithreading is definitely not an easy problem, but can be somehow easier to tackle using the tools that some of the languages provide. Personally, I do not need to use all of the tools on a daily basis, but I think it is worth knowing that they exist and how they can help.