I have a LockManager that manages the locks of several threads. Sometimes the threads are bad boys, and I have to kill them and ask the LockManager to release all their lock
You've discovered a major reason why common wisdom says: Don't kill threads!
Locks are only one of the potential resource leaks that can happen if you forcibly kill a thread. Consider open files and sockets, etc.
Also consider that if you did manage to unlock the lock, there was a reason the lock was locked in the first place. For example, the thread may have partially updated a data structure, and allowing access to that structure from another thread is likely to cause strange and wondrous program failures that are difficult if not impossible to debug.
The best way to handle this situation is to ask the thread to go away. Add a "stop()" method to the object associated with the thread (you do have an object for each thread, don't you?) that sets a flag, and have the thread check this flag regularly and exit if it is set.
If your threads are misbehaving in a way that prevents them from checking the stop flag, then the correct approach is to fix the code so that it does not misbehave.
I've done this with an AtomicReference which gets zero points for elegance, but I don't know of another way.
class PseudoLock {
private final AtomicReference<Boolean> mylock = new AtomicReference<>(Boolean.FALSE);
boolean trylock() {
return mylock.compareAndSet(Boolean.FALSE, Boolean.TRUE);
}
void unlock() {
boolean done = mylock.compareAndSet(Boolean.TRUE, Boolean.FALSE);
if (!done) {
throw new IllegalStateException("Cannot unlock an unlocked thread");
}
}
}
''''
As stated in the comments, killing threads is not a good practice. Most frameworks do their best to interrupt threads in worker queues, but they will only have an effect if the executed code checks the interrupt flag, either by calling Thread.isInterrupted()
or calling an interruptable IO or lock method.
If you really need the concept of killing the execution of code, have a look at the Process
class. You can create a Process
by calling Runtime.exec()
or using the ProcessBuilder
. Calling Process.destroyForcibly()
will forcibly terminate the running process.
Would you be allowed to use your own Lock
? Here's a class that completely proxies the Lock
but when it is told to force the unlock it merely replaces the lock it is proxying with a new one. This should have the effect you want. Sadly it still does not deal with the locks that are left dangling but that now becomes somebody else's problem. Your locks are now magically unlocked.
static class LockProxy<L extends Lock> implements Lock {
// The actual lock.
private volatile Lock lock;
public LockProxy(L lock) {
// Trap the lock we are proxying.
this.lock = lock;
}
@Override
public void lock() {
// Proxy it.
lock.lock();
}
@Override
public void lockInterruptibly() throws InterruptedException {
// Proxy it.
lock.lockInterruptibly();
}
@Override
public boolean tryLock() {
// Proxy it.
return lock.tryLock();
}
@Override
public boolean tryLock(long l, TimeUnit tu) throws InterruptedException {
// Proxy it.
return lock.tryLock(l, tu);
}
@Override
public void unlock() {
// Proxy it.
lock.unlock();
}
@Override
public Condition newCondition() {
// Proxy it.
return lock.newCondition();
}
// Extra functionality to unlock from any thread.
public void forceUnlock() {
// Actually just replace the perhaps locked lock with a new one.
// Kinda like a clone. I expect a neater way is around somewhere.
if (lock instanceof ReentrantLock) {
lock = new ReentrantLock();
} else {
throw new UnsupportedOperationException(
"Cannot force unlock of lock type "
+ lock.getClass().getSimpleName());
}
}
}
Why don't you simply wrap the code of your thread around the following:
ReentrantLock lock = ... obtain your lock somehow ...
lock.lock();
try {
... the "bad boy" code here ...
} finally {
lock.unlock();
}
Then, when your thread finishes (either by normally finishing, or by throwing an exception from your "kill"), it will release the lock.
This is actually the way Oracle recommends using ReentrantLock: http://docs.oracle.com/javase/7/docs/api/java/util/concurrent/locks/ReentrantLock.html