p. The JIRA team has been doing some work to properly support Crowd integration. One of the things we needed to do was to get some of a user’s details from Crowd (name, email address) and some from JIRA. The problem is, JIRA uses a numeric ID to access properties, but Crowd just gives us a user name. So, we need to map the user name to an ID.
p. The first time this happens we need to create a new IDName mapping entry in the database, but it is quite possible, even quite likely in some circumstances that this may happen concurrently. We could simply try and insert them into the database and wait for duplicate key exceptions, but they aren’t all that easy to decode across all the databases we support – besides, it feels dirty.
p. Dushan and Dylan (D&D) were mulling over some solutions with maps of locks and other hairy schemes when they realised that what they really wanted was a map of concurrent operations where each operation will return a value based on the key. They came up with the following interface:
bc.. /**
* This will allow you to submit an operation, encapsulated by a
* {@link Callable}, and keyed by an Object , such that the result of the
* Callable will be available to any concurrent callers with the same Object
* key. The result of the Callable is discarded after the operation has
* succeeded.

* Essentially, this is a Map whose elements expire very quickly. It is
* particularly useful for situations where you need to get or create the
* result, and you want to prevent concurrent creation of the result.
public interface ConcurrentOperationMap
* Returns a value that must be computed via an operation. Implementations
* MUST guarantee that no two operations with the same key may be run
* concurrently. It is up to the implementation to decide whether these must
* be run serially, or if the operation is idempotent to borrow the result
* of the currently executing operation.
* @param key
* the key, like any map key this should be an immutable object
* that correctly implements {@link #hashCode()} and
* {@link #equals(Object)}
* @param operation
* is the operation to execute whose result will be accessible to
* any concurrent callers with the same key.
* @return result of the operation
* @throws ExecutionException
* if the callable generated a checked exception, otherwise a
* runtime exception or error will be thrown
R runOperation(K key, Callable operation) throws ExecutionException;
p. You submit the operation (a ["Callable":http://java.sun.com/j2se/1.5.0/docs/api/java/util/concurrent/Callable.html]) that does the heavy lifting, and a key. If another thread comes along the implementation should either block and wait before running (if the operation is not ["idempotent":http://en.wikipedia.org/wiki/Idempotence_(computer_science)]) or block and borrow the result of the current operation if it is.
p. For our purposes the operation is idempotent so D&D implemented one that borrows the result:
bc.. public class BorrowingConcurrentOperationMap implements
private final ConcurrentMap<K, FutureTask> map
= new ConcurrentHashMap<K, FutureTask>();
public R runOperation(K key, Callable operation)
throws ExecutionException
FutureTask task = map.get(key);
while (task == null)
map.putIfAbsent(key, new FutureTask(operation));
task = map.get(key);
R result = runAndGet(task);
return result;
R runAndGet(FutureTask task) throws ExecutionException
// Concurrent calls to run do not matter as run will be a
// no-op if already running
return task.get();
catch (VariousExceptions handleTheseNicely)

p. Any concurrent calls to @runOperation()@ will get the current FutureTask. If we are unlucky and someone puts one in in between, we fail on the @map.putIfAbsent(…)@ and if we are unlucky and someone removes it in between that failure and the get we loop and put a new @FutureTask@ into the map.
p. The whole thing effectively operates as a kind of high eviction rate cache. It could potentially slash the number of expensive lookup database queries when they get hot-spotted.
p. Its pretty simple – but kind of elegant – I like it!
p. PS. Particular thumbs to Doug Lea et al. for the awesome util.concurrent stuff and the “backport”:http://dcl.mathcs.emory.edu/util/backport-util-concurrent/ that means we can use it in 1.4. Also if you haven’t read “Java Concurrency in Practice (Brian Goetz et al.)”:http://jcip.net/ yet, do – its awesome.