Font Awesome Free 5.13.0 by @fontawesome - License - (Icons: CC BY 4.0, Fonts: SIL OFL 1.1, Code: MIT License)

Thread safety (Java)

Concurrency issues in Java and how to prevent them

Concurrency issues

Even if you don't use low-level concurrency mechanisms, you still need to be careful about concurrency issues!

The issues below can for example still arise when using parallel streams, if lambdas passed to filter() or map() use shared data

Concurrency issue: visibility

Basic idea: updates to a variable might not be visible across different threads!


private static boolean done = false;
public static void main(String[] args) {
ExecutorService executor = Executors.newCachedThreadPool();
executor.execute(() -> {
for (int i = 0; i < 1000; i++) {
System.out.println("Hello " + i);
done = true;
executor.execute(() -> {
int i = 0;
while (!done) {
System.out.println("Goodbye " + i);

Result you might expect: first all "Hello" messages, then a "Goodbye" message with a high number

Result you are likely to get: all "Hello" messages, and then the program never prints "Goodbye" but also never completes

Problem: the effect of done = true; in the thread printing the "Hello" messages might not be visible to the to thread that needs to print the "Goodbye" message

Cause: optimizations performed by compilers, the JVM and processors

  • Processor tries to cache values from RAM in much faster processor cache
  • The order of instructions might be changed in an attempt to improve performance
    • As the inner part of the while loop in the code above doesn't change done, this code might be changed to if (!done) while (true) i++;
  • These optimizations assume that there is no concurrent memory access

Ways to ensure that an update to a variable is visible:

  • The value of a final variable is visible after initialization
    • Good practice: make variables final if possible
  • The initial value of a static variable is visible after static initialization
    • Static initialization: code running in a static {} block
  • Changes to a volatile variable are visible
  • Changes that happen before releasing a lock are visible to anyone acquiring the same lock afterwards (see Locking)

In the example above, marking the variable done as volatile fixes the problem

Concurrency issue: race conditions

A race condition is a situation where multiple threads are each attempting to perform an operation and the interaction between the threads yields an incorrect result that is different from the result that we would get if we just applied each thread's operation sequentially

Example race condition:

private static volatile int count = 0;
public static void main(String[] args) {
ExecutorService executor = Executors.newCachedThreadPool();
for (int i = 0; i < 100; i++) {
int taskId = i;
executor.execute(() -> {
for (int j = 0; j < 1000; j++) {
System.out.println(taskId + " " + count);

Result you might expect: task 999 printing the number 100000

Result you are likely to get: task 999 printing a number lower than 100000, and a different one every time you run the program

Problem: count++ is not an atomic operation! It is actually equivalent to the following:

temp = count + 1;
// at this point, another thread might take over and compute and/or set the new count
count = temp; // if something happened in between, this likely sets an incorrect value

Counters are definitely not the only problem. Race conditions can lead to issues whenever variables shared between threads are mutated. In fact, a lot of Java data structures (for example ArrayList and HashSet) are not thread-safe and can become corrupted if they are accessed from multiple threads.

Strategies for safe concurrency

  • Confinement
    • Don't share any data between threads
    • Data from different threads can be combined after they have finished their computations
  • Immutability
    • Immutable objects are safe to share
      • Of course, overwriting a shared variable holding an immutable object can still lead to issues! (see also the concurrency issue above, taking into account that an integer value is somewhat equivalent to an immutable object)
    • Examples of immutable classes: String and the classes from the Date and Time API
    • Example: it's safe share a single LocalDate object among multiple threads, because all operations on the object (like adding a year to it) do not change the object itself but return a new object instead
    • Tips for implementing your own immutable classes
      • Make all instances variables final (this also helps with visibility, see above)
      • Don't leak any state that could be mutated externally (example: don't return a reference to an internal array or collection from any of your methods). Return a copy instead.
      • Don't store a reference to a mutable object received in the constructor. Make a copy of the object instead.
      • Don't let the this reference escape the constructor (or someone could observe the object in an incomplete state)
  • Thread-safe data structures
    • Some data structures are intended to be used concurrently by multiple threads (see below)
  • Locking
    • Low-level concurrency mechanism
    • Can be used to ensure that a sequence of operations is carried out without being interrupted
      • This is called a critical section
    • Is used internally by thread-safe classes to control concurrency
    • Very hard to get right, so avoid implementing manually when possible
      • Locks can become bottlenecks or even lead to deadlocks
      • Inadequate locking might still allow concurrency issues to occur
    • For more details, see Locking

Thread-safe data structures

Thread-safe collections

Collections in java.util.concurrent are thread-safe and efficient: multiple threads can access them concurrently without corrupting them and the threads won't even block each other if they access different parts

Example: ConcurrentHashMap

ConcurrentHashMap<String, Long> map = new ConcurrentHashMap<>();
// thread-safe methods
map.putIfAbsent("key", 1L);
map.compute("key", (key, value) -> value == null ? 1 : value + 1);
// don't do this, this is not thread-safe
Long oldValue = map.get("key");
Long newValue = oldValue == null ? 1 : oldValue + 1);
map.put("key", newValue);

Note: you can get a concurrent Set (that is internally backed by a ConcurrentHashMap) using ConcurrentHashMap.newKeySet()

Useful for coordinating work between tasks: blocking queues

  • Examples: LinkedBlockingQueue and ArrayBlockingQueue
  • Producer tasks add elements into the queue, consumer tasks retrieve them
  • The queue blocks when trying to add an element through put() if the queue is currently full
    • this makes producer tasks block if they get too far ahead of the consumers
  • The queue blocks when trying to retrieve an element through take()if the queue is empty
    • this makes consumer tasks block until the producers catch up
  • In addition to the blocking methods
    • Methods that throw on queue full/empty: add(), remove(), element()
    • Methods that return null false on queue full/empty: offer(), poll(), peak()
  • Challenge: stopping the consumers
    • An empty queue doesn't necessarily mean that the work is over, it might just mean the producers need to catch up
    • If there is a singe producer, it can help to add a "last item" indicator after the last real item in the queue

Copy-on-write collections: CopyOnWriteArrayList and CopyOnWriteArraySet

  • All mutators make a copy of the underlying array
  • Useful if there are a lot more threads reading the collection than threads mutating it
  • If the collection is mutated after creation of an iterator, the iterator still refers to the old array
    • Iterator has a consistent view, but it might be outdated

Atomic counters and accumulators

Package java.util.concurrent.atomic has some useful classes that provide a thread-safe way to create counters and accumulators


AtomicLong nextId = new AtomicLong();
// in some thread
long id = nextId.incrementAndGet(); // gets value, adds 1, sets it and returns it

The incrementAndGet() method is atomic, which means that it executes all of its operations before another thread is able to access the AtomicLong. This means that multiple threads can safely access the same AtomicLong instance concurrently.

AtomicLong largest = new AtomicLong();
// thread-safe
largest.updateAndGet(x -> Math.max(x, observed));
// not thread-safe
largest.set(Math.max(largest.get(), observed));

Drawback: updates are performed optimistically

  • The operation first computes the new value, then checks if the old value is still the same, and retries if it's not
  • This might not work well with a large number of threads all trying to make updates, as there will be a lot of retries required

Better alternative when lots of threads will be updating at the same time: LongAdder

  • Keeps multiple variables, the sum of all those variables makes up the current value
  • This is efficient in the common situation where we only need the sum after all the work has been done
LongAdder count = new LongAdder();
// in some thread

Thread-local variables

Sometimes, you might want multiple threads to have access to an instance of a certain non-tread-safe class, but you don't need every thread to use the exact same instance. In this case, you can avoid sharing between the threads by giving every thread its own specific instance of the class.

Example: letting multiple threads "share" a NumberFormat

private static final ThreadLocal<NumberFormat> currencyFormat = ThreadLocal
.withInitial(() -> NumberFormat.getCurrencyInstance());
// inside some thread
String amountString = currencyFormat.get().format(total)

The first time a thread calls the get() method, an instance is created based on the lambda expression supplied to withInitial(). From then on, that same instance will always be returned whenever that specific thread calls get().


  • Core Java SE 9 for the Impatient (book by Cay S. Horstmann)