Lock-free atomic update to immutable Map

问题

Given a Javaslang / Vavr immutable map, and a function that updates that map:

private Map<Foo, Bar> myMap = HashMap.empty();

public void setBar(Foo foo, Bar bar) {
  myMap = myMap.put(foo, bar);
}

How can I ensure that two concurrent calls to setBar() for different Foo keys will both have their updates recorded?

// thread A
setBar(fooA, barA)

// thread B
setBar(fooB, barB)

It seems like there's a risk that the calls will be interleaved such that:

1. thread A gets {}
2. thread B gets {}
3. thread B computes {} + fooB -> barB = {(fooB -> barB)}
4. thread B sets myMap to {(fooB -> barB)}
5. thread A computes {} + fooA -> barA = {(fooA -> barA)}
6. thread A sets myMap to {(fooA -> barA)}
7. thread B's update is lost

Using AtomicReference, I came up with the following, more or less based on the ConcurrentStack methods in the “Nonblocking Algorithms” section of Java Concurrency in Practice.

private AtomicReference<Map<Foo, Bar>> myMap = 
  new AtomicReference<>(HashMap.empty());

public void setBar(Foo foo, Bar bar) {
  Map<Foo, Bar> myMap0;
  Map<Foo, Bar> myMap1;
  do {
    myMap0 = myMap.get();
    myMap1 = myMap0.put(foo, bar);
  } while (!myMap.compareAndSet(myMap0, myMap1));
}

Is this correct? And if so, is it as good an implementation as I'm likely to get, or is there something simpler (e.g. some Java 8 AtomicReference API I'm missing that implements this pattern)?

回答1:

Using AtomicReference is good in this case. You can use the shortcut method

public void setBar(Foo foo, Bar bar) {
    myMap.updateAndGet(map -> map.put(foo, bar)));
}

instead. See the javadoc for AtomicReference.updateAndGet. The default java implementation is exactly the same as yours in Java 8.

来源：https://stackoverflow.com/questions/44147719/lock-free-atomic-update-to-immutable-map