ChangeLogTag: Mon Mar 9 17:30:45 UTC 2009 Friedhelm Wolf <fwolf@dre.vanderbilt.edu>

5 files changed, 38 insertions, 8 deletions

diff --git a/TAO/ChangeLog b/TAO/ChangeLog
index 50f884c9afb..7244d266f6b 100644
--- a/TAO/ChangeLog
+++ b/TAO/ChangeLog
@@ -1,3 +1,16 @@
+Mon Mar  9 17:30:45 UTC 2009  Friedhelm Wolf  <fwolf@dre.vanderbilt.edu>
+
+        * orbsvcs/orbsvcs/LWFT/AppSideMonitor_Thread.cpp
+        * orbsvcs/orbsvcs/LWFT/Failure_Handler.cpp
+        * orbsvcs/orbsvcs/LWFT/HostMonitorImpl.cpp:
+        
+          Changed debugging output.
+        
+        * orbsvcs/orbsvcs/LWFT/ReplicationManager.cpp:
+        
+          Added functionality to remove state sync agents in case of
+          process failures.
+
 Mon Mar  9 15:54:19 UTC 2009  Friedhelm Wolf  <fwolf@dre.vanderbilt.edu>
 
         * MPC/config/lwft_server.mpb
diff --git a/TAO/orbsvcs/orbsvcs/LWFT/AppSideMonitor_Thread.cpp b/TAO/orbsvcs/orbsvcs/LWFT/AppSideMonitor_Thread.cpp
index 01342589ab5..c19b1978d11 100644
--- a/TAO/orbsvcs/orbsvcs/LWFT/AppSideMonitor_Thread.cpp
+++ b/TAO/orbsvcs/orbsvcs/LWFT/AppSideMonitor_Thread.cpp
@@ -111,9 +111,9 @@ AppSideMonitor_Thread::svc (void)
           port_ = AppOptions::instance ()->port ();
         }
 
-//      ACE_DEBUG ((LM_TRACE,
-//                  "AppSideReg::svc - got heartbeat port "
-//                  "%d from hm.\n", port_));
+      ACE_DEBUG ((LM_TRACE,
+                  "AppSideReg::svc - got heartbeat port "
+                  "%d from hm.\n", port_));
 
       if (serv_addr_.set (port_) == -1)
         {
@@ -140,13 +140,15 @@ AppSideMonitor_Thread::svc (void)
     }
   }
 
-  //ACE_DEBUG ((LM_DEBUG, "Entering reactor event loop.\n"));
+  ACE_DEBUG ((LM_TRACE, "AppSideMonitor_Thread: Entering reactor event loop.\n"));
   if (reactor_.run_reactor_event_loop() == -1)
     {
       ACE_ERROR_RETURN ((LM_ERROR,
                          "run_reactor_event_loop failed\n"),
                         -1);
     }
+
+  ACE_DEBUG ((LM_TRACE, "AppSideMonitor_Thread: finishing.\n"));
   
   return 0;
 }
diff --git a/TAO/orbsvcs/orbsvcs/LWFT/Failure_Handler.cpp b/TAO/orbsvcs/orbsvcs/LWFT/Failure_Handler.cpp
index f2afe3aa075..802b96776e3 100644
--- a/TAO/orbsvcs/orbsvcs/LWFT/Failure_Handler.cpp
+++ b/TAO/orbsvcs/orbsvcs/LWFT/Failure_Handler.cpp
@@ -31,7 +31,7 @@ int Failure_Handler::handle_input (ACE_HANDLE fd)
   
   if (process_map_.find (fd, pinfo) == 0) /// if found
   {
-    ACE_DEBUG ((LM_DEBUG,
+    ACE_DEBUG ((LM_TRACE,
                 "It looks like process %s has failed.\n",
                 pinfo.process_id.c_str()));
     process_map_.unbind (fd);
diff --git a/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.cpp b/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.cpp
index 1f22e174b60..09e1a051284 100644
--- a/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.cpp
+++ b/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.cpp
@@ -61,13 +61,13 @@ HostMonitorImpl::register_process (const char *process_id,
                           hostname,
                           port);
   process_map_.bind (process_id, handler);
-  /*
+
   ACE_DEBUG ((LM_TRACE,
               "HostMonitorImpl::register_process "
               "process_id = %s, port = %d.\n",
               process_id,
               port));
-  */
+
   return true;
 }
 
diff --git a/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.cpp b/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.cpp
index ae18770d2ce..9184ec2a78e 100644
--- a/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.cpp
+++ b/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.cpp
@@ -631,7 +631,7 @@ ReplicationManager_i::process_proc_failure (
                 {
                   ids[index++] = (*it).c_str ();
                 }
-              // for now just take the first entry and send it
+
               send_failure_notice (host.c_str (),
                                    ids);
             }
@@ -654,6 +654,21 @@ ReplicationManager_i::process_proc_failure (
                           "Data structure invariant broken.\n",
                           host.c_str()));
             }
+
+          {
+            // a failed process means that the state synchronization
+            // agent is no longer reachable either.
+            ACE_Guard <ACE_Thread_Mutex> guard (state_sync_agent_list_mutex_);
+
+            // remove state synchronization agent reference
+            if (state_synchronization_agent_map_.unbind (process_id) == 0)
+              {
+                ACE_DEBUG ((LM_TRACE,
+                            "RM: process_proc_failure () : "
+                            "removed state sync agent from list.\n"));
+              }
+          }
+
         }
       else
         {