Improved handling of marking processes as dead

Run ndb_mgmd as deamon
Make extra attempt to check if processes are still alive

mysql-test/lib/mtr_process.pl

@@ -272,40 +272,17 @@ sub spawn_parent_impl {
           last;
         }
 
-        # If one of the mysqld processes died, we want to
+        # If one of the processes died, we want to
         # mark this, and kill the mysqltest process.
 
-        foreach my $idx (0..1)
-        {
-          if ( $::master->[$idx]->{'pid'} eq $ret_pid )
-          {
-            mtr_debug("child $ret_pid was master[$idx], " .
-                      "exit during mysqltest run");
-            $::master->[$idx]->{'pid'}= 0;
-            last;
-          }
-        }
-
-        foreach my $idx (0..2)
-        {
-          if ( $::slave->[$idx]->{'pid'} eq $ret_pid )
-          {
-            mtr_debug("child $ret_pid was slave[$idx], " .
-                      "exit during mysqltest run");
-            $::slave->[$idx]->{'pid'}= 0;
-            last;
-          }
-        }
-
-        mtr_debug("waitpid() caught exit of unknown child $ret_pid, " .
-                  "exit during mysqltest run");
+	mark_process_dead($ret_pid);
       }
 
       if ( $ret_pid != $pid )
       {
         # We terminated the waiting because a "mysqld" process died.
         # Kill the mysqltest process.
-
+	mtr_verbose("Kill mysqltest because another process died");
         kill(9,$pid);
 
         $ret_pid= waitpid($pid,0);
@@ -639,15 +616,21 @@ sub mtr_check_stop_servers ($) {
               mtr_warning("couldn't delete $file");
             }
           }
+	  $srv->{'pid'}= 0;
         }
       }
     }
     if ( $errors )
     {
-      # We are in trouble, just die....
+      # There where errors killing processes
+      # do one last attempt to ping the servers
+      # and if they can't be pinged, assume they are dead
+      if ( ! mtr_ping_with_timeout( \@$spec ) )
+      {
 	mtr_error("we could not kill or clean up all processes");
       }
     }
+  }
 
   # FIXME We just assume they are all dead, for Cygwin we are not
   # really sure
@@ -773,39 +756,23 @@ sub mtr_ping_with_timeout($) {
   return $res;
 }
 
-##############################################################################
+
 #
-#  The operating system will keep information about dead children, 
-#  we read this information here, and if we have records the process
-#  is alive, we mark it as dead.
+# Loop through our list of processes and look for and entry
+# with the provided pid
+# Set the pid of that process to 0 if found
 #
-##############################################################################
-
-sub mtr_record_dead_children () {
-
-  my $ret_pid;
-
-  # Wait without blockinng to see if any processes had died
-  # -1 or 0 means there are no more procesess to wait for
-  while ( ($ret_pid= waitpid(-1,&WNOHANG)) != 0 and $ret_pid != -1)
-  {
-    mtr_warning("waitpid() caught exit of child $ret_pid");
-    foreach my $idx (0..1)
-    {
-      if ( $::master->[$idx]->{'pid'} eq $ret_pid )
-      {
-        mtr_warning("child $ret_pid was master[$idx]");
-        $::master->[$idx]->{'pid'}= 0;
-      }
-    }
+sub mark_process_dead($)
+{
+  my $ret_pid= shift;
 
-    foreach my $idx (0..2)
+  foreach my $mysqld (@{$::master}, @{$::slave})
   {
-      if ( $::slave->[$idx]->{'pid'} eq $ret_pid )
+    if ( $mysqld->{'pid'} eq $ret_pid )
     {
-        mtr_warning("child $ret_pid was slave[$idx]");
-        $::slave->[$idx]->{'pid'}= 0;
-        last;
+      mtr_verbose("$mysqld->{'type'} $mysqld->{'idx'} exited, pid: $ret_pid");
+      $mysqld->{'pid'}= 0;
+      return;
     }
   }
 
@@ -813,21 +780,43 @@ sub mtr_record_dead_children () {
   {
     if ( $cluster->{'pid'} eq $ret_pid )
     {
-       mtr_warning("child $ret_pid was $cluster->{'name'} cluster ndb_mgmd");
+      mtr_verbose("$cluster->{'name'} cluster ndb_mgmd exited, pid: $ret_pid");
       $cluster->{'pid'}= 0;
-       last;
+      return;
     }
 
     foreach my $ndbd (@{$cluster->{'ndbds'}})
     {
       if ( $ndbd->{'pid'} eq $ret_pid )
       {
-	 mtr_warning("child $ret_pid was $cluster->{'name'} cluster ndbd");
+	mtr_verbose("$cluster->{'name'} cluster ndbd exited, pid: $ret_pid");
 	$ndbd->{'pid'}= 0;
-	 last;
+	return;
       }
     }
   }
+  mtr_warning("mark_process_dead couldn't find an entry for pid: $ret_pid");
+
+}
+
+##############################################################################
+#
+#  The operating system will keep information about dead children, 
+#  we read this information here, and if we have records the process
+#  is alive, we mark it as dead.
+#
+##############################################################################
+
+sub mtr_record_dead_children () {
+
+  my $ret_pid;
+
+  # Wait without blockinng to see if any processes had died
+  # -1 or 0 means there are no more procesess to wait for
+  while ( ($ret_pid= waitpid(-1,&WNOHANG)) != 0 and $ret_pid != -1)
+  {
+    mtr_warning("mtr_record_dead_children: $ret_pid");
+    mark_process_dead($ret_pid);
   }
 }
 
@@ -843,7 +832,8 @@ sub start_reap_all {
   my $pid;
   while(($pid= waitpid(-1, &WNOHANG)) != 0 and $pid != -1)
   {
-    print "start_reap_all: pid: $pid.\n";
+    mtr_warning("start_reap_all pid: $pid");
+    mark_process_dead($pid);
   };
 }
 
@@ -903,6 +893,7 @@ sub sleep_until_file_created ($$$) {
     # Check if it died after the fork() was successful
     if ( $pid != 0 && waitpid($pid,&WNOHANG) == $pid )
     {
+      mtr_warning("Process $pid died");
       return 0;
     }
 

mysql-test/mysql-test-run.pl

@@ -1687,6 +1687,7 @@ sub ndbcluster_wait_started($){
 }
 
 
+
 sub mysqld_wait_started($){
   my $mysqld= shift;
 
@@ -1706,6 +1707,7 @@ sub ndb_mgmd_start ($) {
   mtr_init_args(\$args);
   mtr_add_arg($args, "--no-defaults");
   mtr_add_arg($args, "--core");
+  mtr_add_arg($args, "--nodaemon");
   mtr_add_arg($args, "--config-file=%s", "$cluster->{'data_dir'}/config.ini");
 
 
@@ -1716,9 +1718,23 @@ sub ndb_mgmd_start ($) {
 		  "",
 		  { append_log_file => 1 });
 
+
+  # FIXME Should not be needed
+  # Unfortunately the cluster nodes will fail to start
+  # if ndb_mgmd has not started properly
+  sleep(1);
+
+ # if (!sleep_until_file_created($cluster->{'path_pid'},
+ #				30, # Seconds
+ #				$pid))
+ #  {
+ #    mtr_warning("Failed to start ndb_mgd for $cluster->{'name'} cluster");
+ #    return 1;
+ #  }
+
   # Remember pid of ndb_mgmd
   $cluster->{'pid'}= $pid;
-  mtr_verbose("ndb_mgmd_start, pid: $pid");
+
   return $pid;
 }
 
@@ -1774,19 +1790,6 @@ sub ndbcluster_start ($$) {
 
   my $pid= ndb_mgmd_start($cluster);
 
-  # FIXME Should not be needed
-  # Unfortunately cluster will fail
-  # if ndb_mgmd has not started properly
-  # Wait for the ndb_mgmd pid file to be created
-  if (!sleep_until_file_created($cluster->{'path_pid'},
-				60,
-				$pid))
-  {
-    mtr_warning("Failed to start ndb_mgmd for $cluster->{'name'} cluster");
-    return 1;
-  }
-
-
   for ( my $idx= 0; $idx < $cluster->{'nodes'}; $idx++ )
   {
     ndbd_start($cluster, $idx, $extra_args);