kapok

Contents: What is a thread-local variable? Implement a per-thread Singleton Simplify debug logging InheritableThreadLocal ThreadLocal performance The benefits of ThreadLocal Resources About the author Rate this article Related content: Writing multithreaded Java applications Connection pools Subscriptions: dW newsletters dW Subscription (CDs and downloads)

Brian Goetz (brian@quiotix.com)
Software Consultant, Quiotix Corp.
16 Oct 2001

The ThreadLocal class appeared with little fanfare in version 1.2 of the Java platform. While support for thread-local variables has long been a part of many threading facilities, such as the Posix pthreads facility, the initial design of the Java Threads API lacked this useful feature. Further, the initial implementation was quite inefficient. For these reasons, ThreadLocal gets relatively little attention, but it can be very handy for simplifying the development of thread-safe concurrent programs. In this third installment of Threading lightly, Java software consultant Brian Goetz examines ThreadLocal and offers tips for exploiting its power.
Participate in Brian's Multithreaded Java programming discussion forum for help with threading and concurrency issues in your projects.

Writing thread-safe classes is difficult. It requires a careful analysis of not only the conditions under which variables will be read or written, but also of how the class might be used by other classes. Sometimes, it is very difficult to make a class thread-safe without compromising its functionality, ease of use, or performance. Some classes retain state information from one method invocation to the next, and it is difficult to make such classes thread-safe in any practical way.

It may be easier to manage the use of a non-thread-safe class than to try and make the class thread-safe. A class that is not thread-safe can often be used safely in a multithreaded program as long as you ensure that instances of that class used by one thread are not used by other threads. For example, the JDBC Connection class is not thread-safe -- two threads cannot safely share a Connection at a fine level of granularity -- but if each thread had its own Connection, then multiple threads can safely perform database operations simultaneously.

It is certainly possible to maintain a separate JDBC connection (or any other object) for each thread without the use of ThreadLocal; the Thread API gives us all the tools we need to associate objects with threads. However, the ThreadLocal class makes it much easier for us to manage the process of associating a thread with its per-thread data.

What is a thread-local variable?
A thread-local variable effectively provides a separate copy of its value for each thread that uses it. Each thread can see only the value associated with that thread, and is unaware that other threads may be using or modifying their own copies. Some compilers (such as the Microsoft Visual C++ compiler or the IBM XL FORTRAN compiler) have incorporated support for thread-local variables into the language using a storage-class modifier (like static or volatile). Java compilers offer no special language support for thread-local variables; instead, they are implemented with the ThreadLocal class, which has special support in the core Thread class.

Because thread-local variables are implemented through a class, rather than as part of the Java language itself, the syntax for using thread-local variables is a bit more clumsy than for language dialects where thread-local variables are built in. To create a thread-local variable, you instantiate an object of class ThreadLocal. The ThreadLocal class behaves much like the various Reference classes in java.lang.ref; it acts as an indirect handle for storing or retrieving a value. Listing 1 shows the ThreadLocal interface.

public class ThreadLocal { 
  public Object get();
  public void set(Object newValue);
  public Object initialValue();
}

The get() accessor retrieves the current thread's value of the variable; the set() accessor modifies the current thread's value. The initialValue() method is an optional method that lets you set the initial value of the variable if it has not yet been used in this thread; it allows for a form of lazy initialization. How ThreadLocal behaves is best illustrated by an example implementation. Listing 2 shows one way to implement ThreadLocal. It isn't a particularly good implementation (although it is quite similar to the initial implementation), as it would likely perform poorly, but it illustrates clearly how ThreadLocal behaves.

public class ThreadLocal { 
  private Map values = Collections.synchronizedMap(new HashMap());

  public Object get() {
    Thread curThread = Thread.currentThread();
    Object o = values.get(curThread);
    if (o == null && !values.containsKey(curThread)) {
      o = initialValue();
      values.put(curThread, o);
    }
    return o;
  }

  public void set(Object newValue) {
    values.put(Thread.currentThread(), newValue);
  }

  public Object initialValue() {
    return null;
  }
}

This implementation will perform poorly because it requires synchronization on the values map for each get() and set() operation, and if multiple threads are accessing the same ThreadLocal at once, there will be contention. Additionally, this implementation is impractical because using Thread objects as the key in the values map will prevent the Thread from being garbage collected after the thread exits, and the thread-specific values of the ThreadLocal for deceased threads will also not be garbage collected.

Using ThreadLocal to implement a per-thread Singleton
Thread-local variables are commonly used to render stateful Singleton or shared objects thread-safe, either by encapsulating the entire unsafe object in a ThreadLocal or by encapsulating the object's thread-specific state in a ThreadLocal. For example, in an application that is tightly tied to a database, many methods may need to access the database. It could be inconvenient to include a Connection as an argument to every method in the system -- a sloppier, but significantly more convenient technique would be to access the connection with a Singleton. However, multiple threads cannot safely share a JDBC Connection. By using a ThreadLocal in our Singleton, as shown in Listing 3, we can allow any class in our program to easily acquire a reference to a per-thread Connection. In this way, we can think of a ThreadLocal as allowing us to create a per-thread-singleton.

public class ConnectionDispenser { 
  private static class ThreadLocalConnection extends ThreadLocal {
    public Object initialValue() {
      return DriverManager.getConnection(ConfigurationSingleton.getDbUrl());
    }
  }

  private ThreadLocalConnection conn = new ThreadLocalConnection();

  public static Connection getConnection() {
    return (Connection) conn.get();
  }
}

Any stateful or non-thread-safe object that is relatively more expensive to create than to use, such as a JDBC Connection or a regular-expression matcher, is a good candidate for the per-thread-singleton technique. Of course, for situations like this, you can use other approaches, like pooling, for safely managing shared access. However, even pooling has some potential drawbacks from a scalability perspective. Because pool implementations must synchronize to maintain the integrity of the pool data structures, if all threads are using the same pool, program performance may suffer due to contention in a system with many threads accessing the pool frequently.

Using ThreadLocal to simplify debug logging
Other applications for ThreadLocal in which pooling would not be a useful alternative include storing or accumulating per-thread context information for later retrieval. For example, suppose you wanted to create a facility for managing debugging information in a multithreaded application. You could accumulate debugging information in a thread-local container as shown by the DebugLogger class in Listing 4. At the beginning of a unit of work, you empty the container, and when an error occurs, you query the container to retrieve all the debugging information that has been generated so far by this unit of work.

public class DebugLogger {
  private static class ThreadLocalList extends ThreadLocal {
    public Object initialValue() {
      return new ArrayList();
    }

    public List getList() { 
      return (List) super.get(); 
    }
  }

  private ThreadLocalList list = new ThreadLocalList();
  private static String[] stringArray = new String[0];

  public void clear() {
    list.getList().clear();
  }

  public void put(String text) {
    list.getList().add(text);
  }

  public String[] get() {
    return list.getList().toArray(stringArray);
  }
}

Throughout your code, you can call DebugLogger.put(), saving information about what your program is doing, and you can easily retrieve the debugging information relevant to a particular thread later when necessary (such as when an error has occurred). This technique is a lot more convenient and efficient than simply dumping everything to a log file and then trying to sort out which log records come from which thread (and worrying about contention for the logging object between threads.)

ThreadLocal is also useful in servlet-based applications or any multithreaded server application in which the unit of work is an entire request, because then a single thread will be used during the entire course of handling the request. You can use ThreadLocal variables to store any sort of per-request context information using the per-thread-singleton technique described earlier.

ThreadLocal's less thread-safe cousin, InheritableThreadLocal
The ThreadLocal class has a relative, InheritableThreadLocal, which functions in a similar manner, but is suitable for an entirely different sort of application. When a thread is created, if it holds values for any InheritableThreadLocal objects, these values are automatically passed on to the child process as well. If a child process calls get() on an InheritableThreadLocal, it sees the same object as the parent would. To preserve thread-safety, you should use InheritableThreadLocal only for immutable objects (objects whose state will not ever be changed once created), because the object is shared between multiple threads. InheritableThreadLocal is useful for passing data from a parent thread to a child thread, such as a user id, or a transaction id, but not for stateful objects like JDBC Connections.

ThreadLocal performance

Don't miss the rest of this series Part 1, "Synchronization is not the enemy" (July 2001) Part 2, "Reducing contention" (September 2001)

While the concept of a thread-local variable has been around for a long time and is supported by many threading frameworks including the Posix pthreads specification, thread-local support was omitted from the initial Java Threads design and only added in version 1.2 of the Java platform. In many ways, ThreadLocal is still a work in progress; it was rewritten for version 1.3 and again for version 1.4, both times to address performance problems.

In JDK 1.2, ThreadLocal was implemented in a manner very similar to Listing 2, except that a synchronized WeakHashMap was used to store the values instead of a HashMap. (Using WeakHashMap solves the problem of Thread objects not getting garbage collected, at some additional performance cost.) Needless to say, the performance of ThreadLocal was quite poor.

The version of ThreadLocal provided with version 1.3 of the Java platform is substantially better; it does not use any synchronization and so does not present a scalability problem, and it does not use weak references either. Instead, the Thread class was modified to support ThreadLocal by adding an instance variable to Thread that holds a HashMap mapping thread-local variables to their values for the current thread. Because the process of retrieving or setting a thread-local variable does not involve reading or writing data that might be read or written by another thread, you can implement ThreadLocal.get() and set() without any synchronization. Also, because the references to the per-thread values are stored in the owning Thread object, when the Thread gets garbage collected, so can its per-thread values.

Unfortunately, even with these improvements, the performance of ThreadLocal under Java 1.3 is still surprisingly slow. My rough benchmarks running the Sun 1.3 JDK on a two-processor Linux system show that a ThreadLocal.get() operation takes about twice as long as an uncontended synchronization. The reason for this poor performance is that the Thread.currentThread() method is quite expensive, accounting for more than two-thirds of the ThreadLocal.get() run time. Even with these weaknesses, the JDK 1.3 ThreadLocal.get() is still much faster than a contended synchronization, so if there is any significant chance of contention at all (perhaps there is a large number of threads, or the synchronized block is executed frequently, or the synchronized block is large), ThreadLocal may still be more efficient overall.

Under the newest version of the Java platform, version 1.4b2, performance of ThreadLocal and Thread.currentThread() has been improved significantly. With these new improvements, ThreadLocal should be faster than other techniques such as pooling. Because it is simpler and often less error-prone than those other techniques, it will eventually be discovered as an effective way to prevent undesired interactions between threads.

The benefits of ThreadLocal
ThreadLocal offers a number of benefits. It is often the easiest way to render a stateful class thread-safe, or to encapsulate non-thread-safe classes so that they can safely be used in multithreaded environments. Using ThreadLocal allows us to bypass the complexity of determining when to synchronize in order to achieve thread-safety, and it improves scalability because it doesn't require any synchronization. In addition to simplicity, using ThreadLocal to store a per-thread-singleton or per-thread context information has a valuable documentation perk -- by using a ThreadLocal, it's clear that the object stored in the ThreadLocal is not shared between threads, simplifying the task of determining whether a class is thread-safe or not.

I hope you've enjoyed and learned from this series, and I encourage you to follow up on your multithreading issues in my discussion forum.

Resources

• Participate in the discussion forum on this article. (You can also click Discuss at the top or bottom of the article to access the forum.)
  
  
• Java Performance Tuning by Jack Shirazi (O'Reilly & Associates, 2000) provides guidance on eliminating performance issues in the Java platform.
  
  
• Java Platform Performance: Strategies and Tactics by Steve Wilson and Jeff Kesselman (Addison-Wesley, 2000) offers techniques for the experienced Java programmer to build speedy and efficient Java code.
  
  
• Java Performance and Scalability, Volume 1: Server-Side Programming Techniques by Dov Bulka (Addison-Wesley, 2000) provides a wealth of tips and tricks designed to help you increase the performance of your apps.
  
  
• Brian Goetz' article "Double-checked locking: Clever, but broken" (JavaWorld, February 2001) explores the Java Memory Model (JMM) in detail and the surprising consequences of failing to synchronize in certain situations.
  
  
• Doug Lea's Concurrent Programming in Java, Second Edition (Addison-Wesley, 1999) is a masterful book on the subtle issues surrounding multithreaded programming in the Java language.
  
  
• In his article "Writing multithreaded Java applications" (developerWorks, February 2001), Alex Roetter introduces the Java Thread API, outlines issues involved in multithreading, and offers solutions to common problems.
  
  
• "Connection pools" (developerWorks, October 2000) by Siva Visverwaran focuses on support for connection pooling of both database resources and non-database resources in a J2EE environment.
  
  
• This article, part of an IBM Redbook, describes how optimistic locking in WebSphere lets different transactions read the same state concurrently, and checks data integrity only at update time.
  
  
• The performance modeling and analysis team at IBM Thomas J. Watson Research Center is researching several projects in the areas of performance and performance management.
  
  
• Find more Java technology resources in the developerWorks Java technology zone.