ChangeLogTag: Sun Mar 22 03:04:48 UTC 2009 Friedhelm Wolf <fwolf@dre.vanderbilt.edu>

8 files changed, 89 insertions, 17 deletions

diff --git a/TAO/ChangeLog b/TAO/ChangeLog
index 94c48321cda..287612e1b89 100644
--- a/TAO/ChangeLog
+++ b/TAO/ChangeLog
@@ -1,3 +1,15 @@
+Sun Mar 22 03:04:48 UTC 2009  Friedhelm Wolf  <fwolf@dre.vanderbilt.edu>
+
+        * orbsvcs/orbsvcs/LWFT/ReplicationManager.h
+        * orbsvcs/orbsvcs/LWFT/HMOptions.h
+        * orbsvcs/orbsvcs/LWFT/HostMonitorImpl.h
+        * orbsvcs/orbsvcs/LWFT/ReplicationManager.cpp
+        * orbsvcs/orbsvcs/LWFT/HostMonitorImpl.cpp
+        * orbsvcs/orbsvcs/LWFT/HMOptions.cpp
+        * orbsvcs/LWFT_Service/host_monitor.cpp:
+        
+          Added logging for failure timing.
+
 Wed Mar 18 21:06:23 UTC 2009  Friedhelm Wolf  <fwolf@dre.vanderbilt.edu>
 
         * orbsvcs/orbsvcs/LWFT/ReplicationManager.cpp:
diff --git a/TAO/orbsvcs/LWFT_Service/host_monitor.cpp b/TAO/orbsvcs/LWFT_Service/host_monitor.cpp
index 66cfd22ad29..7db1842bc96 100644
--- a/TAO/orbsvcs/LWFT_Service/host_monitor.cpp
+++ b/TAO/orbsvcs/LWFT_Service/host_monitor.cpp
@@ -47,7 +47,7 @@ int main (int argc, char* argv[])
       
       /// Initilize the timedate object on heap.
       HostMonitorImpl * host_monitor =
-        new HostMonitorImpl (orb, &monitor_thread);
+        new HostMonitorImpl (orb, &monitor_thread, opts->logging ());
       PortableServer::ServantBase_var safe_host (host_monitor);
       ACE_UNUSED_ARG (safe_host);
       
diff --git a/TAO/orbsvcs/orbsvcs/LWFT/HMOptions.cpp b/TAO/orbsvcs/orbsvcs/LWFT/HMOptions.cpp
index f95a9116300..2ba72b64783 100644
--- a/TAO/orbsvcs/orbsvcs/LWFT/HMOptions.cpp
+++ b/TAO/orbsvcs/orbsvcs/LWFT/HMOptions.cpp
@@ -2,7 +2,7 @@
 // $Id$
 
 #include <sstream>
-
+#include <iostream>
 #include "ace/Global_Macros.h"
 #include "ace/Guard_T.h"
 #include "ace/Log_Msg.h"
@@ -21,7 +21,8 @@ HMOptions::HMOptions (void)
     port_range_begin_ (7000),
     RM_update_freq_ (1),
     load_monitor_freq_(2),
-    debug_level_ (0)    
+    debug_level_ (0),
+    logging_ (false)
 {
   char hostname [100];
   gethostname (hostname, sizeof (hostname));
@@ -65,6 +66,11 @@ HMOptions::parse_args (int &argc, char **argv)
           HM_ior_file_ = arg;
           as.consume_arg ();
         }
+      else if (0 != (arg = as.get_the_parameter (ACE_TEXT ("-logging"))))
+        {
+          logging_ = true;
+          as.consume_arg ();
+        }
       else if (0 != (arg = as.get_the_parameter (ACE_TEXT ("-rm_ior"))))
         {
           RM_ior_ = arg;
@@ -222,3 +228,9 @@ HMOptions::set_debug_level (void)
   ACE_LOG_MSG->priority_mask (mask,
                               ACE_Log_Msg::PROCESS);
 }
+
+bool
+HMOptions::logging (void)
+{
+  return logging_;
+}
diff --git a/TAO/orbsvcs/orbsvcs/LWFT/HMOptions.h b/TAO/orbsvcs/orbsvcs/LWFT/HMOptions.h
index 4cc93ea5c36..220006dbeca 100644
--- a/TAO/orbsvcs/orbsvcs/LWFT/HMOptions.h
+++ b/TAO/orbsvcs/orbsvcs/LWFT/HMOptions.h
@@ -44,6 +44,7 @@ public:
   std::string util_file (void) const;
   std::pair <char, std::string> ior_access (void) const;
   bool use_naming_service (void) const;
+  bool logging (void);
 
   /**
    *  @return the debug level of the application:
@@ -71,6 +72,7 @@ protected:
   size_t load_monitor_freq_;
   std::pair <char, std::string> ior_access_;
   long debug_level_;
+  bool logging_;
 
   /// Singleton-related stuff.
   static HMOptions * volatile instance_;
diff --git a/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.cpp b/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.cpp
index 85984b1032e..dffdc3b271d 100644
--- a/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.cpp
+++ b/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.cpp
@@ -8,6 +8,7 @@
 #include "ace/Time_Value.h"
 #include "ace/Synch_Options.h"
 #include "ace/Thread_Mutex.h"
+#include "ace/Date_Time.h"
 
 #include "HostMonitorImpl.h"
 #include "Failure_Handler.h"
@@ -16,11 +17,12 @@
 #include "HMOptions.h"
 #include "Utilization_Monitor.h"
 
-HostMonitorImpl::HostMonitorImpl (CORBA::ORB_ptr orb, Monitor_Thread *mt)
+HostMonitorImpl::HostMonitorImpl (CORBA::ORB_ptr orb, Monitor_Thread *mt, bool logging)
   : monitor_thread_ (mt),
     port_counter_ (HMOptions::instance ()->port_range_begin ()),
     connector_ (monitor_thread_->get_reactor ()),
-    orb_ (CORBA::ORB::_duplicate (orb))
+    orb_ (CORBA::ORB::_duplicate (orb)),
+    logging_ (logging)
 {
   this->create_RM_Proxy ();
 }
@@ -126,6 +128,24 @@ int HostMonitorImpl::drop_process (const std::string &process_id)
   if (process_map_.find (process_id, handler) == 0)
     {
       rm_proxy_->proc_failure (process_id);
+
+      if (logging_)
+        {
+          std::ofstream logfile;
+          logfile.open ("./hm.log", ios_base::app);
+
+          ACE_Date_Time dt;
+
+          logfile << process_id << " " 
+                  << dt.hour () << ":" 
+                  << dt.minute () << ":" 
+                  << dt.second () << ":"
+                  << dt.microsec ()
+                  << std::endl;
+
+          logfile.close ();
+        }
+
       return remove_process (process_id);
     }
   
diff --git a/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.h b/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.h
index a8b31f6bdc8..82ca29078c2 100644
--- a/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.h
+++ b/TAO/orbsvcs/orbsvcs/LWFT/HostMonitorImpl.h
@@ -29,7 +29,7 @@ public:
                            ACE_Thread_Mutex>
     ProcessHandlerMap;
 
-  HostMonitorImpl (CORBA::ORB_ptr, Monitor_Thread *);
+  HostMonitorImpl (CORBA::ORB_ptr, Monitor_Thread *, bool logging = false);
   
   int
   drop_process (const std::string &process_id);
@@ -69,6 +69,7 @@ private:
   ProcessHandlerMap process_map_;
   ACE_Connector <Failure_Handler, ACE_SOCK_Connector> connector_;
   CORBA::ORB_var orb_;
+  bool logging_;
 };
 
 #endif // __HOSTMONITORIMPL_H
diff --git a/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.cpp b/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.cpp
index 347db876543..6ecd8dcd218 100644
--- a/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.cpp
+++ b/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.cpp
@@ -6,7 +6,7 @@
 
 #include <string>
 #include <set>
-
+#include <ace/Date_Time.h>
 #include "ace/OS_NS_unistd.h"
 #include <ace/High_Res_Timer.h>
 #include "AppSideMonitor_Thread.h"
@@ -684,15 +684,26 @@ ReplicationManager_i::process_proc_failure (
   std::ofstream out;
   out.open ("rm-failures.txt", ios_base::app);
 
-  for (PROC_FAIL_TIME_LIST::iterator it = failure_history_.begin ();
-       it != failure_history_.end ();
-       ++it)
-    {
-      out << it->second << " " << it->first << std::endl;
-    }
+  {
+    ACE_Guard <ACE_Thread_Mutex> guard (history_lock_);
+
+    ACE_Date_Time dt;
+    for (PROC_FAIL_TIME_LIST::iterator it = failure_history_.begin ();
+         it != failure_history_.end ();
+         ++it)
+      {
+        dt.update (it->timestamp);
+        out << it->process_id << " " 
+            << dt.hour () << ":"
+            << dt.minute () << ":"
+            << dt.second () << ":"
+            << dt.microsec () << " "
+            << it->delta << std::endl;
+      }
 
-  out.close ();
-  failure_history_.clear ();
+    out.close ();
+    failure_history_.clear ();
+  }
 }
 
 void
@@ -1676,7 +1687,14 @@ ReplicationManager_i::proc_failure (const char *process_id)
 
   ACE_Time_Value tv;
   timer_.elapsed_time (tv);
-  failure_history_.push_back (TProcessFailureTime (tv.msec (), process_id));
+  TProcessFailureTime sample = {tv.msec (), ACE_OS::gettimeofday (), process_id};
+
+  {
+    ACE_Guard <ACE_Thread_Mutex> guard (history_lock_);
+
+    failure_history_.push_back ( sample );
+  }
+
   timer_.reset ();
   timer_.start ();
 }
diff --git a/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.h b/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.h
index 2ebd6751ec6..b9179a4a88d 100644
--- a/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.h
+++ b/TAO/orbsvcs/orbsvcs/LWFT/ReplicationManager.h
@@ -321,10 +321,17 @@ private:
 
   ACE_High_Res_Timer timer_;
   
-  typedef std::pair <unsigned long, std::string> TProcessFailureTime;
+  struct TProcessFailureTime {
+    unsigned long delta;
+    ACE_Time_Value timestamp;
+    std::string process_id;
+  };
+
   typedef std::list <TProcessFailureTime> PROC_FAIL_TIME_LIST;
   PROC_FAIL_TIME_LIST failure_history_;
 
+  ACE_Thread_Mutex history_lock_;
+
   void update_enhanced_ranklist (void);
     
   void update_map (const char * key_str,