Object locking in Distributed cache

Question

Hi,
     I have gone through some of the posts on the forum regarding the locking issues.
     Thread at http://www.tangosol.net/forums/thread.jspa?messageID=3416 specifies that
     ..locks block locks, not gets or puts. If one member locks a key, it prevents another member from locking the key. It does not prevent the other member from getting/putting that key.
     What exactly do we mean by the above statement?
     I'm using distributed cache and would like to lock an object before "put" or "remove" operations. But I want "read" operation to be without locks. Now my questions are,
     1) In a distributed cache setup, if I try to obtain a lock before put or remove and discover that i cannot obtain it (i.e false is returned) because someone else has locked it, then how do I discover when the other entity releases lock and when should i retry for the lock?
     2) Again, if i lock USING "LOCK(OBJECT KEY)" method, can i be sure that no other cluster node would be writing/reading to that node until I release that lock?
     3) The first post in the thread http://www.tangosol.net/forums/thread.jspa?forumID=37&threadID=588 suggests that in distributed setup locks are of no effect to other cluster nodes. But i want locks to be valid across cluster. If item locked, then no one else shud be able to transact with it. How to get that?
     Regards,
     Mohnish

Answer 1

Hi Mohnish,
     >> 1) In a distributed cache setup, if I try to obtain
     >> a lock before put or remove and discover that i cannot
     >> obtain it (i.e false is returned) because someone else
     >> has locked it, then how do I discover when the other
     >> entity releases lock and when should i retry for the
     >> lock?
     You may try to acquire a lock (waiting indefinitely for lock acquisition) by calling <tt>cache.lock(key, -1)</tt>; you may try for a specified time period by calling <tt>cache.lock(key, cMillis)</tt>. With either of these approaches, your thread will block until the other thread releases the lock (or the timeout is reached). In either case, if the other node releases its lock your lock request will complete immediately and your thread will resume execution.
     >> 2) Again, if i lock USING "LOCK(OBJECT KEY)" method,
     >> can i be sure that no other cluster node would be
     >> writing/reading to that node until I release that lock?
     If you want to prevent other threads from writing/reading a cache entry, you must ensure that those other threads lock the key prior to writing/reading it. If they do not lock it prior to writing, they may have dirty writes (overwriting your write or vice versa); if they do not lock prior to reading, they may have dirty reads (having the underlying data change after they've read it).
     >> 3) The first post in the thread
     http://www.tangosol.net/forums/thread.jspa?forumID=37&threadID=588
     >> suggests that in distributed setup locks are of no
     >> effect to other cluster nodes. But i want locks to be
     >> valid across cluster. If item locked, then no one else
     >> shud be able to transact with it. How to get that?
     The first post in that thread states that if the second thread doesn't lock, then it will overwrite the first thread (even if the first thread has locked). However, there is an inconsistency in that the Replicated cache topology has a stronger memory model than the Distributed/Near topologies. In the Replicated topoplogy, locks not only block out locks from other nodes, they also prevent puts from other nodes.
     Jon Purdy
     Tangosol, Inc.