From 17b94163f0b092b9b66bcb334ab2258ab8f806b1 Mon Sep 17 00:00:00 2001
From: mellor <mellor@8638fb3e-16cb-4fca-ae20-7b5d299a9bcd>
Date: Mon, 5 Jan 2009 18:02:00 +0000
Subject: [PATCH] minor changes for demo

git-svn-id: https://svn.code.sf.net/p/xcat/code/xcat-core/trunk@2559 8638fb3e-16cb-4fca-ae20-7b5d299a9bcd
---
 xCAT-server/lib/xcat/plugins/rollupdate.pm | 33 ++++++++++++++--------
 xCAT-server/share/xcat/rollupdate/ll.tmpl  |  6 +++-
 2 files changed, 26 insertions(+), 13 deletions(-)

diff --git a/xCAT-server/lib/xcat/plugins/rollupdate.pm b/xCAT-server/lib/xcat/plugins/rollupdate.pm
index cc7906f54..d11c4b897 100644
--- a/xCAT-server/lib/xcat/plugins/rollupdate.pm
+++ b/xCAT-server/lib/xcat/plugins/rollupdate.pm
@@ -654,6 +654,18 @@ sub ll_jobs {
             close($JOBFILE);
             chown( $uid, $gid, $lljob_file );
 
+			# Need to change status before actually submittly LL jobs
+			# If LL jobs happen to run right away, the update code checking
+			# for the status may run before we've had a chance to actually update it
+            my $nltab = xCAT::Table->new('nodelist');
+            my @nodes = split( /\,/, $nodelist );
+            $nltab->setNodesAttribs(
+                                     \@nodes,
+                                     {
+                                        appstatus =>
+                                          "ROLLUPDATE-update_job_submitted"
+                                     }
+            );
             # Submit LL job
             my $cmd = qq~su - $lluser "-c llsubmit $lljob_file"~;
 			if ($::VERBOSE) {
@@ -669,16 +681,6 @@ sub ll_jobs {
                 xCAT::MsgUtils->message( "E", $rsp, $::CALLBACK );
                 return 1;
             }
-
-            my $nltab = xCAT::Table->new('nodelist');
-            my @nodes = split( /\,/, $nodelist );
-            $nltab->setNodesAttribs(
-                                     \@nodes,
-                                     {
-                                        appstatus =>
-                                          "ROLLUPDATE-update_job_submitted"
-                                     }
-            );
         }
         elsif ( defined($nodelist) ) {
 
@@ -826,6 +828,9 @@ sub rebootnodes {
         }
     }
 
+    # my $nltab = xCAT::Table->new('nodelist');
+    $nltab->setNodesAttribs( \@nodes, { appstatus => "ROLLUPDATE-shutting_down" } );
+
     # remove nodes from LL
     $scheduler =~ tr/[A-Z]/[a-z]/;
     if ( $scheduler eq 'loadleveler' ) {
@@ -852,7 +857,8 @@ sub rebootnodes {
             if ( defined( $machines{$node} )
                  && ( $machines{$node}{'mstatus'} eq "1" ) )
             {
-                my $cmd = "llctl -h $node drain";
+                #my $cmd = "llctl -h $node drain";
+                my $cmd = "llctl -h $node flush startd";
 				if ($::VERBOSE) { 
 					open (RULOG, ">>$::LOGDIR/$::LOGFILE");
 					print RULOG localtime()." Running command \'$cmd\'\n";
@@ -861,6 +867,8 @@ sub rebootnodes {
                 xCAT::Utils->runcmd( $cmd, 0 );
             }
         }
+		# give LL a chance to catch up
+		sleep 15;
     }
 
     # Shutdown the nodes
@@ -951,7 +959,8 @@ sub rebootnodes {
     $nltab->setNodesAttribs( \@nodes, { appstatus => "ROLLUPDATE-rebooting" } );
     if ( scalar(@rnetboot_nodes) > 0 ) {
         my $rnb_nodelist = join( ',', @rnetboot_nodes );
-        my $cmd = "rnetboot $rnb_nodelist -f";
+        # my $cmd = "rnetboot $rnb_nodelist -f";
+        my $cmd = "rpower $rnb_nodelist on";
 		if ($::VERBOSE) { 
 			open (RULOG, ">>$::LOGDIR/$::LOGFILE");
 			print RULOG localtime()." Running command \'$cmd\' \n";
diff --git a/xCAT-server/share/xcat/rollupdate/ll.tmpl b/xCAT-server/share/xcat/rollupdate/ll.tmpl
index ad5373dbf..73f24fd4c 100644
--- a/xCAT-server/share/xcat/rollupdate/ll.tmpl
+++ b/xCAT-server/share/xcat/rollupdate/ll.tmpl
@@ -14,14 +14,18 @@
 #
 # @ job_name = rollupdate_[[NODESET]]
 # @ job_type = parallel
+## Note:  really want node usage to be not shared, but there
+##   was a timing bug in LL that sometimes caused a rollupdate job to 
+##   not start.  Should be fixed by now.
 # @ node_usage = not_shared
 # @ restart = no
 # @ error   = /home/lluser/rollupdate_jobs/rollupdate_[[NODESET]].$(Host).$(Cluster).$(Process).err
 # @ output  = /home/lluser/rollupdate_jobs/rollupdate_[[NODESET]].$(Host).$(Cluster).$(Process).out
 # @ executable = /opt/xcat/share/xcat/rollupdate/send_reboot_request
-# @ arguments = 5 loadleveler [[XCATSERVER]] [[XCATPORT]] [[XNODELIST]] 
+# @ arguments = 10 --verbose loadleveler [[XCATSERVER]] [[XCATPORT]] [[XNODELIST]] 
 # @ node = [[LLCOUNT]]
 # @ tasks_per_node = 1
 # @ requirements = (Machine == {[[LLMACHINES]]}) 
+# @ wall_clock_limit = 21:00,20:00
 # @ queue