ConcurrentModificationException despite using synchronized
public synchronized X getAnotherX(){
if(iterator.hasNext()){
X b = iterator.next();
String name = b.getInputFileName();
...
return b;
}
else{return null;}
}
despite the synchronized statement in the declaration header, i still get a ConcurrentModificationException Exception at the line where i use iterator.next(); whats wrong here ?
ConcurrentModificationException
usually has nothing to do with multiple threads. Most of the time it occurs because you are modifying the collection over which it is iterating within the body of the iteration loop. For example, this will cause it:
Iterator iterator = collection.iterator();
while (iterator.hasNext()) {
Item item = (Item) iterator.next();
if (item.satisfiesCondition()) {
collection.remove(item);
}
}
In this case you must use the iterator.remove()
method instead. This occurs equally if you are adding to the collection, in which case there is no general solution. However, the subtype ListIterator
can be used if dealing with a list and this has an add()
method.
I agree with the statements above about ConcurrentModificationException
often happening as a result of modifying the collection in the same thread as iterating. However, it is not always the reason.
The thing to remember about synchronized
is that it only guarantees exclusive access if everybody accessing the shared resource also synchronizes.
For example, you can synchronize access to a shared variable:
synchronized (foo) {
foo.setBar();
}
And you can think that you have exclusive access to it. However, there is nothing to stop another thread just doing something without the synchronized
block:
foo.setBar(); // No synchronization first.
Through bad luck (or Murphy's Law, "Anything that can go wrong, will go wrong."), these two threads can happen to execute at the same time. In the case of structural modifications of some widely-used collections (e.g. ArrayList
, HashSet
, HashMap
etc), this can result in a ConcurrentModificationException
.
It is hard to prevent the problem entirely:
-
You can document synchronization requirements, e.g. inserting "you must synchronize on
blah
before modifying this collection" or "acquirebloo
lock first", but that's relying upon users to discover, read, understand and apply the instruction.There is the
javax.annotation.concurrent.GuardedBy
annotation, which can help to document this in a standardized way; the problem is then that you have to have some means of checking correct use of the annotation in the toolchain. For example, you might be able to use something like Google's errorprone, which can check in some situations, but it's not perfect. -
For simple operations on collections, you can make use of the
Collections.synchronizedXXX
factory methods, which wrap a collection so that every method call synchronizes on the underlying collection first, e.g. theSynchronizedCollection.add
method:@Override public boolean add(E e) { synchronized (mutex) { return c.add(obj); } }
Where
mutex
is the synchronized-on instance (often theSynchronizedCollection
itself), andc
is the wrapped collection.The two caveats with this approach are:
-
You have to be careful that the wrapped collection cannot be accessed in any other way, since that would allow non-synchronized access, the original problem. This is typically achieved by wrapping the collection immediately on construction:
Collections.synchronizedList(new ArrayList<T>());
-
The synchronization is applied per method call, so if you are doing some compound operation, e.g.
if (c.size() > 5) { c.add(new Frob()); }
then you don't have exclusive access throughout that operation, only for the
size()
andadd(...)
calls individually.In order to get mutually exclusive access for the duration of the compound operation, you would need to externally synchronize, e.g.
synchronized (c) { ... }
. This requires you to know the correct thing to synchronize on, however, which may or may not bec
.
-