Wed May 1 21:06:50 UTC 2013 Phil Mesnier <mesnier_p@ociweb.com>

6 files changed, 206 insertions, 50 deletions

diff --git a/TAO/ChangeLog b/TAO/ChangeLog
index 4637326ab77..339a60c43b5 100644
--- a/TAO/ChangeLog
+++ b/TAO/ChangeLog
@@ -1,9 +1,29 @@
+Wed May  1 21:06:50 UTC 2013  Phil Mesnier  <mesnier_p@ociweb.com>
+
+        * orbsvcs/ImplRepo_Service/ImR_Locator_i.cpp:
+
+          Repopulate the pinger with all known servers on a restart with
+          a persistent store. This helps the list -a command return
+          accurate results. Also list -a considers all final states that
+          are not LS_DEAD to be be equivalent to alive.
+
+        * orbsvcs/ImplRepo_Service/LiveCheck.h:
+        * orbsvcs/ImplRepo_Service/LiveCheck.cpp:
+
+          Ensure that when a server is in a final state of LS_LAST_TRANSIENT
+          a new listener will be notified of that state.
+
+        * orbsvcs/tests/ImplRepo/servers_interact_on_startup/README:
+        * orbsvcs/tests/ImplRepo/servers_interact_on_startup/run_test.pl:
+
+          Add new test variants for listing and locator restarts.
+
 Wed May  1 14:16:54 UTC 2013  Phil Mesnier  <mesnier_p@ociweb.com>
 
         * orbsvcs/tests/ImplRepo/servers_interact_on_startup/run_test.pl:
 
-          Clear the result returned by the tao_imr when it is expected to return a
-          non-zero value.
+          Clear the result returned by the tao_imr when it is expected to
+          return a non-zero value.
 
 Wed May  1 13:32:09 UTC 2013  Phil Mesnier  <mesnier_p@ociweb.com>
 
diff --git a/TAO/orbsvcs/ImplRepo_Service/ImR_Locator_i.cpp b/TAO/orbsvcs/ImplRepo_Service/ImR_Locator_i.cpp
index 5eb152c0434..835939619a4 100644
--- a/TAO/orbsvcs/ImplRepo_Service/ImR_Locator_i.cpp
+++ b/TAO/orbsvcs/ImplRepo_Service/ImR_Locator_i.cpp
@@ -164,11 +164,40 @@ ImR_Locator_i::init_with_orb (CORBA::ORB_ptr orb, Options& opts)
   ACE_ASSERT (! CORBA::is_nil (ior_table.in ()));
   ior_table->set_locator (this->ins_locator_.in ());
 
+
   // initialize the repository. This will load any values that
   // may have been persisted before.
-  return this->repository_->init(this->root_poa_.in (),
-                                 this->imr_poa_.in (),
-                                 ior);
+  int result = this->repository_->init(this->root_poa_.in (),
+                                       this->imr_poa_.in (),
+                                       ior);
+  if (result == 0)
+    {
+      Locator_Repository::SIMap::ENTRY* entry = 0;
+      Locator_Repository::SIMap::ITERATOR it (this->repository_->servers ());
+
+      // Number of servers that will go into the server_list.
+      CORBA::ULong n = this->repository_->servers ().current_size ();
+
+      for (CORBA::ULong i = 0; i < n; i++)
+        {
+          it.next (entry);
+          it.advance ();
+          ACE_ASSERT (entry != 0);
+
+          const Server_Info& info = *(entry->int_id_);
+          //          ImplementationRepository::ServerInformation_var imr_info =
+          //            info.createImRServerInfo ();
+
+          if (!this->pinger_.has_server (info.name.c_str()))
+            {
+              this->pinger_.add_server (info.name.c_str(),
+                                        this->ping_external_,
+                                        info.server.in());
+            }
+        }
+    }
+
+  return result;
 }
 
 int
@@ -1289,6 +1318,12 @@ ImR_Locator_i::connect_server (UpdateableServerInfo& info)
 {
   if (! CORBA::is_nil (info->server.in ()))
     {
+      if (!this->pinger_.has_server (info->name.c_str()))
+        {
+          this->pinger_.add_server (info->name.c_str(),
+                                    this->ping_external_,
+                                    info->server.in());
+        }
       return; // already connected
     }
 
@@ -1323,6 +1358,11 @@ ImR_Locator_i::connect_server (UpdateableServerInfo& info)
         ORBSVCS_DEBUG ((LM_DEBUG,
                     ACE_TEXT ("ImR: Connected to server <%C>\n"),
                     info->name.c_str ()));
+      this->pinger_.add_server (info->name.c_str(),
+                                this->ping_external_,
+                                info->server.in());
+
+
     }
   catch (CORBA::Exception&)
     {
@@ -1435,7 +1475,7 @@ SyncListener::is_alive (void)
       this->orb_->perform_work (delay);
     }
   this->got_it_ = false;
-  return this->status_ == LS_ALIVE;
+  return this->status_ != LS_DEAD;
 }
 
 bool
diff --git a/TAO/orbsvcs/ImplRepo_Service/LiveCheck.cpp b/TAO/orbsvcs/ImplRepo_Service/LiveCheck.cpp
index 087d91666d8..76ef51a3ebc 100644
--- a/TAO/orbsvcs/ImplRepo_Service/LiveCheck.cpp
+++ b/TAO/orbsvcs/ImplRepo_Service/LiveCheck.cpp
@@ -214,7 +214,6 @@ LiveEntry::LiveEntry (LiveCheck *owner,
     ref_ (ImplementationRepository::ServerObject::_duplicate (ref)),
     liveliness_ (LS_UNKNOWN),
     next_check_ (ACE_High_Res_Timer::gettimeofday_hr()),
-    retry_count_ (0),
     repings_ (0),
     max_retry_ (LiveEntry::reping_limit_),
     may_ping_ (may_ping),
@@ -239,9 +238,10 @@ void
 LiveEntry::reset_status (void)
 {
   ACE_GUARD (TAO_SYNCH_MUTEX, mon, this->lock_);
-  if ( this->liveliness_ == LS_ALIVE)
+  if ( this->liveliness_ == LS_ALIVE || this->liveliness_ == LS_LAST_TRANSIENT)
     {
       this->liveliness_ = LS_UNKNOWN;
+      this->repings_ = 0;
       this->next_check_ = ACE_High_Res_Timer::gettimeofday_hr();
     }
 }
@@ -265,29 +265,14 @@ LiveEntry::status (void) const
 }
 
 void
-LiveEntry::status (LiveStatus l)
+LiveEntry::update_listeners (void)
 {
-  {
-    ACE_GUARD (TAO_SYNCH_MUTEX, mon, this->lock_);
-    this->liveliness_ = l;
-    if (l == LS_ALIVE)
-      {
-        this->retry_count_ = 0;
-        ACE_Time_Value now (ACE_High_Res_Timer::gettimeofday_hr());
-        this->next_check_ = now + owner_->ping_interval();
-      }
-  }
   Listen_Set remove;
-
-  LiveStatus ls = this->liveliness_;
-  if (ls == LS_TRANSIENT && ! this->reping_available())
-    ls = LS_LAST_TRANSIENT;
-
   for (Listen_Set::ITERATOR i(this->listeners_);
        !i.done();
        i.advance ())
     {
-      if ((*i)->status_changed (ls))
+      if ((*i)->status_changed (this->liveliness_))
         {
           remove.insert (*i);
         }
@@ -302,6 +287,25 @@ LiveEntry::status (LiveStatus l)
         this->listeners_.remove (*i);
       }
   }
+}
+
+void
+LiveEntry::status (LiveStatus l)
+{
+  {
+    ACE_GUARD (TAO_SYNCH_MUTEX, mon, this->lock_);
+    this->liveliness_ = l;
+    if (l == LS_ALIVE)
+      {
+        ACE_Time_Value now (ACE_High_Res_Timer::gettimeofday_hr());
+        this->next_check_ = now + owner_->ping_interval();
+      }
+    if (l == LS_TRANSIENT && !this->reping_available())
+      {
+        this->liveliness_ = LS_LAST_TRANSIENT;
+      }
+  }
+  this->update_listeners ();
 
   if (this->listeners_.size() > 0)
     {
@@ -328,22 +332,17 @@ LiveEntry::next_check (void) const
 bool
 LiveEntry::validate_ping (bool &want_reping, ACE_Time_Value& next)
 {
-  if (ImR_Locator_i::debug () > 2)
-    {
-      ORBSVCS_DEBUG ((LM_DEBUG,
-                      ACE_TEXT ("(%P|%t) LiveEntry::validate_ping, server = %C\n"),
-                      this->server_.c_str()));
-    }
   if (this->liveliness_ == LS_PING_AWAY ||
       this->liveliness_ == LS_DEAD ||
       this->listeners_.size () == 0)
     {
-      if (ImR_Locator_i::debug () > 2)
+      if (ImR_Locator_i::debug () > 5)
         {
           ORBSVCS_DEBUG ((LM_DEBUG,
-                          ACE_TEXT ("(%P|%t) LiveEntry::validate_ping, first test, ")
-                          ACE_TEXT ("status = %d, listeners = %d\n"),
-                          this->liveliness_, this->listeners_.size ()));
+                          ACE_TEXT ("(%P|%t) LiveEntry::validate_ping, status ")
+                          ACE_TEXT ("= %d, listeners = %d server %C\n"),
+                          this->liveliness_, this->listeners_.size (),
+                          this->server_.c_str()));
         }
       return false;
     }
@@ -360,11 +359,13 @@ LiveEntry::validate_ping (bool &want_reping, ACE_Time_Value& next)
       if (ImR_Locator_i::debug () > 2)
         {
           ORBSVCS_DEBUG ((LM_DEBUG,
-                          ACE_TEXT ("(%P|%t) LiveEntry::validate_ping, second test, ")
+                          ACE_TEXT ("(%P|%t) LiveEntry::validate_ping, ")
                           ACE_TEXT ("status = %d, listeners = %d, ")
-                          ACE_TEXT ("diff = %d,%d, msec = %d\n"),
+                          ACE_TEXT ("diff = %d,%d, msec = %d ")
+                          ACE_TEXT ("server %C\n"),
                           this->liveliness_, this->listeners_.size (),
-                          diff.sec(), diff.usec(), msec));
+                          diff.sec(), diff.usec(), msec,
+                          this->server_.c_str()));
         }
       return false;
     }
@@ -374,23 +375,43 @@ LiveEntry::validate_ping (bool &want_reping, ACE_Time_Value& next)
       break;
     case LS_ALIVE:
     case LS_TIMEDOUT:
-      this->next_check_ = now + owner_->ping_interval();
+      {
+        ACE_GUARD_RETURN (TAO_SYNCH_MUTEX, mon, this->lock_, false);
+        this->next_check_ = now + owner_->ping_interval();
+      }
       break;
     case LS_TRANSIENT:
+    case LS_LAST_TRANSIENT:
       {
         int ms = this->next_reping ();
         if (ms != -1)
           {
+            ACE_GUARD_RETURN (TAO_SYNCH_MUTEX, mon, this->lock_, false);
+            if (this->liveliness_ == LS_LAST_TRANSIENT)
+              {
+                this->liveliness_ = LS_TRANSIENT;
+              }
             ACE_Time_Value next (ms / 1000, (ms % 1000) * 1000);
             this->next_check_ = now + next;
           }
         else
           {
+            if (this->liveliness_ == LS_TRANSIENT)
+              {
+                ACE_GUARD_RETURN (TAO_SYNCH_MUTEX, mon, this->lock_, false);
+                this->liveliness_ = LS_LAST_TRANSIENT;
+              }
             if (ImR_Locator_i::debug () > 2)
               {
                 ORBSVCS_DEBUG ((LM_DEBUG,
                                 ACE_TEXT ("(%P|%t) LiveEntry::validate_ping, ")
-                                ACE_TEXT ("transient, no more repings\n")));
+                                ACE_TEXT ("transient, no more repings, ")
+                                ACE_TEXT ("server %C\n"),
+                                this->server_.c_str()));
+              }
+            if (this->listeners_.size() > 0)
+              {
+                this->update_listeners ();
               }
             return false;
           }
@@ -414,12 +435,10 @@ LiveEntry::do_ping (PortableServer::POA_ptr poa)
       this->ref_->sendc_ping (cb.in());
       ACE_GUARD (TAO_SYNCH_MUTEX, mon, this->lock_);
       this->liveliness_ = LS_PING_AWAY;
-      this->retry_count_++;
     }
   catch (CORBA::Exception &)
     {
       this->status (LS_DEAD);
-      this->retry_count_++;
     }
 }
 
@@ -568,7 +587,7 @@ LiveCheck::handle_timeout (const ACE_Time_Value &,
         }
       else
         {
-          if (ImR_Locator_i::debug () > 2)
+          if (ImR_Locator_i::debug () > 5)
             {
               ORBSVCS_DEBUG ((LM_DEBUG,
                               ACE_TEXT ("(%P|%t) LiveCheck::handle_timeout(%d)")
@@ -617,6 +636,15 @@ LiveCheck::handle_timeout (const ACE_Time_Value &,
   return 0;
 }
 
+bool
+LiveCheck::has_server (const char *server)
+{
+  ACE_CString s (server);
+  LiveEntry *entry = 0;
+  int result = entry_map_.find (s, entry);
+  return (result == 0 && entry != 0);
+}
+
 void
 LiveCheck::add_server (const char *server,
                        bool may_ping,
diff --git a/TAO/orbsvcs/ImplRepo_Service/LiveCheck.h b/TAO/orbsvcs/ImplRepo_Service/LiveCheck.h
index e975cf1a293..cc8b060e52b 100644
--- a/TAO/orbsvcs/ImplRepo_Service/LiveCheck.h
+++ b/TAO/orbsvcs/ImplRepo_Service/LiveCheck.h
@@ -138,6 +138,7 @@ class Locator_Export LiveEntry
   void status (LiveStatus l);
   void reset_status (void);
 
+  void update_listeners (void);
   bool validate_ping (bool &want_reping, ACE_Time_Value &next);
   void do_ping (PortableServer::POA_ptr poa);
   const ACE_Time_Value &next_check (void) const;
@@ -213,6 +214,8 @@ class Locator_Export LiveCheck : public ACE_Event_Handler
   int handle_timeout (const ACE_Time_Value &current_time,
                       const void *act = 0);
 
+  bool has_server (const char *server);
+
   void add_server (const char *server,
                    bool may_ping,
                    ImplementationRepository::ServerObject_ptr ref);
diff --git a/TAO/orbsvcs/tests/ImplRepo/servers_interact_on_startup/README b/TAO/orbsvcs/tests/ImplRepo/servers_interact_on_startup/README
index e88d67ac7fa..ee5ec01375a 100644
--- a/TAO/orbsvcs/tests/ImplRepo/servers_interact_on_startup/README
+++ b/TAO/orbsvcs/tests/ImplRepo/servers_interact_on_startup/README
@@ -71,12 +71,20 @@ captured in ./imr_loc.log
 -debuglog
 All the processes write log output to individual log files
 
+-hide_server
+Use this along with the -imr_start option to test an error condition of a
+registered server being removed from the filesystem. The tao_imr start
+command should recover without hanging.
+
 -imr_start
 Use this option to run a simple test where the tao_imr start command is
 uesd to launch a server.
 This should not time out.
 
--hide_server
-Use this along with the -imr_start option to test an error condition of a
-registered server being removed from the filesystem. The tao_imr start
-command should recover without hanging.
+-list
+Use the tao_imr list command to test. May be used with or without restarting
+the locator.
+
+-restart_loc
+Use to bounce the locator process by itself mid test. Only used by the list
+test for now.
\ No newline at end of file
diff --git a/TAO/orbsvcs/tests/ImplRepo/servers_interact_on_startup/run_test.pl b/TAO/orbsvcs/tests/ImplRepo/servers_interact_on_startup/run_test.pl
index 5a9207033bb..99925e985c1 100755
--- a/TAO/orbsvcs/tests/ImplRepo/servers_interact_on_startup/run_test.pl
+++ b/TAO/orbsvcs/tests/ImplRepo/servers_interact_on_startup/run_test.pl
@@ -19,15 +19,18 @@ my $client_count = 2;
 my $server_reply_delay = 5;
 my $usage = 0;
 my $use_imr_start = 0;
+my $restart_loc = 0;
+my $list_test = 0;
 my $hide_server = 0;
 my $expect_transient = "";
 my $notify_locator = "";
-
 my $debuglog = "";
 my @srvlogfile = ( "", "", "" );
 my @cltlogfile = ( "", "", "" );
 my $actlogfile = "";
 my $loclogfile = "";
+my $back_store_flag = "";
+my $back_store_file = "";
 
 # Ping interval in milliseconds
 my $verification_interval_msecs = 1000;
@@ -62,6 +65,16 @@ if ($#ARGV >= 0) {
         elsif ($ARGV[$i] eq "-imr_start") {
             $use_imr_start = 1;
         }
+        elsif ($ARGV[$i] eq "-list") {
+            $imr_debug_level = 3;
+            $list_test = 1;
+        }
+        elsif ($ARGV[$i] eq "-restart_loc") {
+            $restart_loc = 1;
+            $back_store_flag = "--directory";
+            $back_store_file = ".";
+            $server_reply_delay = 0;
+        }
 	elsif ($ARGV[$i] eq "-v") {
 	    $i++;
 	    $verification_interval_msecs = $ARGV[$i];
@@ -217,6 +230,7 @@ sub init_test
 
     $IMR->Arguments ("-o $imr_imriorfile $refstyle -orbendpoint iiop://:$port ".
 		     "$forward_on_exception_arg ".
+                     "$back_store_flag $back_store_file ".
 		     "-d $imr_debug_level $imr_debug_arg ".
 		     "-v $verification_interval_msecs");
     print ">>> " . $IMR->CommandLine () . "\n";
@@ -363,13 +377,32 @@ sub fini_test
     return $status;
 }
 
+sub restart_imr_locator
+{
+    print_msg ("restarting locator");
+    my $IMR_status = $IMR->TerminateWaitKill ($imr->ProcessStopWaitInterval());
+    if ($IMR_status != 0) {
+        print STDERR "ERROR: IMR returned $IMR_status\n";
+        $status = 1;
+    }
+
+    print ">>> " . $IMR->CommandLine () . "\n";
+    $IMR_status = $IMR->Spawn ();
+
+    if ($IMR_status != 0) {
+        print STDERR "ERROR: ImplRepo Service returned $IMR_status\n";
+        return 1;
+    }
+
+    print "sleeping (5).... ";
+    sleep (5);
+    print "done\n";
+}
 
 sub run_imr_start_test
 {
     init_test ();
 
-    print_msg ("using IMR to start $obj[1]");
-
     $result = run_imr_util ("start $obj[1]");
 
     if ($hide_server == 1 && $result == 4) {
@@ -382,6 +415,27 @@ sub run_imr_start_test
     return $result;
 }
 
+sub run_list_test
+{
+    init_test ();
+    run_imr_util ("start $obj[1]");
+    run_imr_util ("start $obj[0]");
+    run_imr_util ("list -a");
+
+    if ($restart_loc == 1) {
+        restart_imr_locator ();
+        run_imr_util ("list -a");
+    }
+
+    print "sleeping (20).... ";
+    sleep (20);
+    print "done\n";
+
+    run_imr_util ("list -a");
+
+    fini_test ();
+}
+
 sub run_client_activate_test
 {
     init_test ();
@@ -444,6 +498,9 @@ my $ret = 0;
 if ($use_imr_start == 1) {
     $ret = run_imr_start_test ();
 }
+elsif ($list_test == 1) {
+    $ret = run_list_test ();
+}
 else {
     $ret = run_client_activate_test ();
 }