Fri Aug 1 21:01:55 UTC 2014 Phil Mesnier <mesnier_p@ociweb.com>

        * orbsvcs/ImplRepo_Service/AsyncAccessManager.cpp:
        * orbsvcs/ImplRepo_Service/ImR_Activator_i.h:
        * orbsvcs/ImplRepo_Service/ImR_Activator_i.cpp:
        * orbsvcs/ImplRepo_Service/ImR_Locator.idl:
          Address potential race conditions when server starts are
          requested via both the primary and backup ImRs.

        * orbsvcs/tests/ImplRepo/double_start/client.cpp:
        * orbsvcs/tests/ImplRepo/double_start/restart.sh:
        * orbsvcs/tests/ImplRepo/double_start/run_test.pl:
        * orbsvcs/tests/ImplRepo/double_start/server.cpp:
          Refactor the kill test so that it better matches the original
          complaint.

4 files changed, 69 insertions, 35 deletions

diff --git a/TAO/orbsvcs/tests/ImplRepo/double_start/client.cpp b/TAO/orbsvcs/tests/ImplRepo/double_start/client.cpp
index 844db30bfe0..e1568cd4dba 100644
--- a/TAO/orbsvcs/tests/ImplRepo/double_start/client.cpp
+++ b/TAO/orbsvcs/tests/ImplRepo/double_start/client.cpp
@@ -87,23 +87,28 @@ do_restart_test (void)
   obj = set_timeout_policy (obj.in (), ACE_Time_Value (5,0));
   Test_var test = Test::_narrow( obj.in () );
   ACE_ASSERT (!CORBA::is_nil(test.in ()));
-
-  try
+  int attempt = 2;
+  while (attempt > 0)
     {
-      test->arm ();
-    }
-  catch (const CORBA::Exception& ex)
-    {
-      ACE_DEBUG ((LM_DEBUG,
-                  ACE_TEXT ("client caught %C during arm\n"),
-                  ex._name ()));
-      return;
+      try
+        {
+          test->arm ();
+          attempt = 0;
+        }
+      catch (const CORBA::Exception& ex)
+        {
+          attempt--;
+          ACE_DEBUG ((LM_DEBUG,
+                      ACE_TEXT ("client caught %C during arm, retrying\n"),
+                      ex._name ()));
+        }
     }
 
   ACE_DEBUG ((LM_DEBUG,
               ACE_TEXT ("client sleeping %d seconds\n"),
               request_delay_secs));
   ACE_OS::sleep (request_delay_secs);
+
   try
     {
       test->trigger ();
diff --git a/TAO/orbsvcs/tests/ImplRepo/double_start/restart.sh b/TAO/orbsvcs/tests/ImplRepo/double_start/restart.sh
index 2098e3a1eae..4fb892211c4 100755
--- a/TAO/orbsvcs/tests/ImplRepo/double_start/restart.sh
+++ b/TAO/orbsvcs/tests/ImplRepo/double_start/restart.sh
@@ -39,22 +39,22 @@ if [ 0 -ne $reset_flag ]; then
 #        tao_imr shutdown $server_name
 #        status=$?
 #        echo shutdown returned $status
-        echo restart: calling tao_imr start
+        echo restart $$: calling tao_imr start
         tao_imr start $server_name
         status=$?
-        echo restart: tao_imr start returned $status
+        echo restart $$: tao_imr start returned $status
         exit;
     fi
 fi
 
-echo restart: running $exec_cmd
+echo restart $$: running $exec_cmd
 $exec_cmd &
 exec_cmd_pid=$!
-echo "restart: pid is $exec_cmd_pid, waiting"
+echo "restart $$: pid is $exec_cmd_pid, waiting"
 
 wait $exec_cmd_pid
 exec_cmd_status=$?
-echo "restart: status is $exec_cmd_status"
+echo "restart $$: status is $exec_cmd_status"
 
 if [[ 0 != $exec_cmd_status ]]; then
     ./restart.sh -e "$exec_cmd" -c $use_imr -p $$ -r 1 -s $server_name &
diff --git a/TAO/orbsvcs/tests/ImplRepo/double_start/run_test.pl b/TAO/orbsvcs/tests/ImplRepo/double_start/run_test.pl
index 7e26c0e81f5..e0fcc864a6a 100755
--- a/TAO/orbsvcs/tests/ImplRepo/double_start/run_test.pl
+++ b/TAO/orbsvcs/tests/ImplRepo/double_start/run_test.pl
@@ -12,6 +12,7 @@ use PerlACE::TestTarget;
 $status = 0;
 $debuglevel = 0;
 $cltdbg = 1;
+$cltpause = 0;
 $kill = 0;
 $server_pid = 0;
 
@@ -26,6 +27,7 @@ if ($#ARGV >= 0) {
 	}
         elsif ($ARGV[$i] eq '-kill') {
             $kill = 1;
+            $cltpause = 5;
         }
 	else {
             usage();
@@ -201,7 +203,9 @@ sub get_server_pid
     my $pid = 0;
     open (FILE, "server.pid") or die "Can't open server.pid: $!";
     while (<FILE>) {
-      $pid = $_;
+        chomp;
+        $pid = $_;
+        $server_pid = $pid if ($server_pid == 0);
     }
     close FILE;
     return $pid;
@@ -210,7 +214,6 @@ sub get_server_pid
 sub signal_server
 {
     my $sig = shift;
-    $server_pid = get_server_pid () if ($server_pid == 0);
     print "signal $sig to server $server_pid\n";
     kill ($sig, $server_pid);
 }
@@ -220,8 +223,7 @@ sub start_imr
     my $all = shift;
     my $debugbase = "-ORBDebugLevel $debuglevel " .
                     "-ORBVerboseLogging 1 -ORBLogFile ";
-    my $actargs = "-l -o $act_actiorfile $act_initref -ORBListenEndpoints iiop://127.0.0.1:
-";
+    my $actargs = "-d $debuglevel -l -o $act_actiorfile $act_initref -ORBListenEndpoints iiop://127.0.0.1:";
 
     my $imrargs = " -d $debuglevel -i -v 1000 " .
         "--directory . --primary " .
@@ -307,7 +309,7 @@ sub start_imr
     }
 }
 
-sub run_client
+sub launch_client
 {
     if ($srv->GetFile ($srviorfile) == -1) {
         print STDERR "ERROR: cannot retrieve file <$srv_srviorfile>\n";
@@ -318,16 +320,27 @@ sub run_client
         return 1;
     }
 
-    my $args = "-k file://$srviorfile -ORBSvcConf clt.conf";
+    my $args = "-k file://$srviorfile -ORBSvcConf clt.conf -d $cltpause";
     $args .= " -ORBDebuglevel $cltdbg -ORBVerboseLogging 1 -ORBLogFile $cltlogfile" if ($debuglevel > 0);
 
     print "running client $args\n";
 
     $CLT->Arguments ($args);
-    if ($CLT->SpawnWaitKill ($clt->ProcessStartWaitInterval() + 120) == -1) {
+    if ($CLT->Spawn () == -1) {
         print STDERR "ERROR: client failed\n";
         return 1;
     }
+    return 0;
+}
+
+sub run_client
+{
+    if (launch_client () == 0) {
+        if ($CLT->WaitKill ($clt->ProcessStartWaitInterval() + 120) == -1) {
+            print STDERR "ERROR: client failed\n";
+            return 1;
+        }
+    }
 }
 
 sub do_ti_command
@@ -342,7 +355,7 @@ sub do_ti_command
     print "invoking ti cmd $cmd $obj_name $cmdargs\n" if ($debuglevel > 0);
     $TI->Arguments ("$ti_initref $cmd $obj_name $cmdargs");
     $TI_status = $TI->SpawnWaitKill ($ti->ProcessStartWaitInterval());
-    if ($TI_status != 0) {
+    if ($TI_status != 0 && $TI_status != 5) {
         return kill_imr ("tao_imr $cmd $obj_name returned $TI_status");
     }
 }
@@ -375,7 +388,7 @@ sub list_active_servers
 sub kill_primary_test
 {
     print "Running double server start test killing the primary ImR.\n";
-
+    unlink "server.pid";
     my $result = 0;
     my $start_time = time();
 
@@ -394,23 +407,35 @@ sub kill_primary_test
     if (do_ti_command ("start") != 0) {
         return 1;
     }
+    my $firstpid = get_server_pid ();
+    print "first pid = $firstpid, server pid = $server_pid\n";
 
-    list_active_servers ("-v");
+    print "starting client\n";
 
+    if (launch_client () != 0) {
+        return 1;
+    }
+    sleep 2;
     signal_server ("STOP");
     kill_primary ();
-    sleep 2;
+    sleep 4;
+
     start_imr (0);
-    sleep 2;
-    print "starting client\n";
 
-    if (run_client () != 0) {
+    signal_server ("KILL");
+    unlink "server.pid";
+    $server_pid = 0;
+
+    print "******waiting for client exit\n";
+    if ($CLT->WaitKill ($clt->ProcessStartWaitInterval() + 120) == -1) {
+        print STDERR "ERROR: client failed\n";
         return 1;
     }
+    print "******client done\n";
 
-    signal_server ("CONT");
-
+    sleep 4;
     my $final_pid = get_server_pid ();
+    print "first server pid was $server_pid, but now there is $final_pid\n";
     if ($final_pid != $server_pid) {
         print "first server pid was $server_pid, but now there is $final_pid\n";
         $server_pid = $final_pid;
diff --git a/TAO/orbsvcs/tests/ImplRepo/double_start/server.cpp b/TAO/orbsvcs/tests/ImplRepo/double_start/server.cpp
index 7252db67e82..b6d02da96ea 100644
--- a/TAO/orbsvcs/tests/ImplRepo/double_start/server.cpp
+++ b/TAO/orbsvcs/tests/ImplRepo/double_start/server.cpp
@@ -146,14 +146,18 @@ ACE_TMAIN (int argc, ACE_TCHAR *argv[])
     mgr->activate();
 
     ACE_DEBUG ((LM_DEBUG,
-                ACE_TEXT ("Started Server <%C>\n"),
-                poa_name.c_str()));
+                ACE_TEXT ("Started Server <%C> pid = %d\n"),
+                poa_name.c_str(), ACE_OS::getpid ()));
 
     ACE_CString pid_file = "server.pid";
     {
-      ofstream out(pid_file.c_str ());
+      ofstream out(pid_file.c_str (), ios_base::app);
       out << ACE_OS::getpid () << endl;
+      out.close ();
     }
+    ACE_DEBUG ((LM_DEBUG,
+                ACE_TEXT ("Server <%C> pid = %d closed pidfile\n"),
+                poa_name.c_str(), ACE_OS::getpid ()));
 
     if (max_run > 0)
       {
@@ -168,7 +172,7 @@ ACE_TMAIN (int argc, ACE_TCHAR *argv[])
     root_poa->destroy(1,1);
     orb->destroy();
 
-    ACE_OS::unlink (pid_file.c_str ());
+    //    ACE_OS::unlink (pid_file.c_str ());
 
   }
   catch(const CORBA::Exception& ex) {