FileLock Doesn't Work Across Network

Question

I have found a case in which FileLock doesn't work. I'm trying to use FileLock to prevent two instances of the same process to run simultaneously on different machines (or the same machine) on the network. I actually saw two processes, one on each of two Linux machines obtain a lock on the same file at the same time.
Can I be sure that FileLock is calling fcntl and not flock (which wouldn't work over the network)?
Is there any way to fix this or get around it??
Here is the relevant code:
try{
     File lockFile = new File(getLockFilename());
     lockFile.createNewFile();  // Throw out the boolean return value; it doesn't matter if this succeeded.
     FileChannel fc = new FileOutputStream(lockFile).getChannel();
     fileLock = fc.lock();
catch (FileLockInterruptionException flie){
     // Allow this one.
catch (IOException ioe){
        // Report the exception, release the lock if obtained and exit.
// Code executed while the file is locked includes setting up and registering an RMI server object.
try{
     fileLock.release();
catch (IOException ioe){
     logger.warn("Unable to relase file lock:  " + ioe.getClass().getName() + ":  " + ioe.getMessage());
}

Answer 1

I suppose when you say "across network", you mean via
nfs?Yes.
Have each machine check some central point (or all
other nodes) for the existence of a server. Of course
there could be some synchronization problems, so you
might check several times. Even if the remote
interface is successfully exported, have it check one
final time to ensure. To be paranoid, you could have
it check often.We have a strong requirement not to have a central server, which has made coding this a headache. It would be pretty unacceptable to query every machine on our network (most of which probably aren't running any related process at any one time and are probably running other jobs) any time someone wants to start one of these processes on any machine. (There could be hundreds of these processes running at once with different parameters; it's just that no two with the same parameters are allowed to run at once. Each of these will act as a separate RMI server so other processes can communicate with the running process for other purposes besides just determining whether it is running.) A registry can only contain objects on one machine, right? Or can I somehow bind everything on the same host even though that's not where the objects are?? I am already writing the name of the successful host for one parameter set to a file and barring other processes with the same parameters from executing if looking up the process on that host succeeds. This works in most cases, but there is a race condition if two processes start around the same time and both read the file before either has written its host name there. That's why I'm trying to use a file lock on a different file so the read-bind-write sequence happens while the file is locked.