2 Replies Latest reply on May 18, 2005 1:47 AM by akinf

    Synchronous replication does NOT block the caller

    akinf

      Synchronous replication does NOT block the caller as it is described in the tutorial when running two instances of the same application.We need this functionality to support high availability.
      When the application is clustered, all instances other than the first instance to acquire the lock on the object ,should wait for the object
      until it is replicated to all cluster nodes.Is this possible?

        • 1. Re: Synchronous replication does NOT block the caller

          Please desribe in details which case in the tutorial that you have gone through? It should block.

          -Ben

          • 2. Re: Synchronous replication does NOT block the caller
            akinf

            Thank you for your reply
            In the treecache tutorial , on replication section it writes:

            Replication can be synchronous or asynchronous . Use of either one of the options is application dependent. Synchronous replication blocks the caller (e.g. on a put()) until the modifications have been replicated successfully to all nodes in a cluster. ......

            Also jbosscache examples includes a unittest to test concurrent puts

            public void testConcurrentPuts() throws Exception {
            initCaches(TreeCache.REPL_SYNC);
            cache1.setSyncCommitPhase(true);

            Thread t1=new Thread("Thread1") {
            Transaction tx;

            public void run() {
            try {
            tx=beginTransaction();
            cache1.put("/bela/ban", "name", "Bela Ban");
            _pause(2000); // Thread2 will be blocked until we commit
            tx.commit();
            System.out.println("[Thread1] ** LOCK INFO cache1: " + cache1.printLockInfo());
            System.out.println("[Thread1] ** LOCK INFO cache2: " + cache2.printLockInfo());
            }
            catch(Throwable ex) {
            ex.printStackTrace();
            t1_ex=ex;
            }
            }
            };

            Thread t2=new Thread("Thread2") {
            Transaction tx;

            public void run() {
            try {
            _pause(1000); // give Thread1 time to acquire the lock
            tx=beginTransaction();
            System.out.println("[Thread2] ** LOCK INFO cache1: " + cache1.printLockInfo());
            System.out.println("[Thread2] ** LOCK INFO cache2: " + cache2.printLockInfo());
            cache1.put("/bela/ban", "name", "Michelle Ban");
            System.out.println("[Thread2] ** LOCK INFO cache1: " + cache1.printLockInfo());
            System.out.println("[Thread2] ** LOCK INFO cache2: " + cache2.printLockInfo());
            tx.commit();
            System.out.println("[Thread2] ** LOCK INFO cache1: " + cache1.printLockInfo());
            System.out.println("[Thread2] ** LOCK INFO cache2: " + cache2.printLockInfo());
            }
            catch(Throwable ex) {
            ex.printStackTrace();
            t2_ex=ex;
            }
            }
            };

            // Let the game start
            t1.start();
            t2.start();

            // Wait for threads to die
            t1.join();
            t2.join();

            if(t1_ex != null)
            fail("Thread1 failed: " + t1_ex);
            if(t2_ex != null)
            fail("Thread2 failed: " + t2_ex);

            assertEquals("Michelle Ban", cache1.get("/bela/ban", "name"));
            }

            This test is ok when running in the same jvm.ie Thread2 is blocked until
            first thread commits. But we need this behaviour for two threads running in different jvms.Bu it failed.