arif/xcat-core
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

minor changes for demo

Browse Source 
git-svn-id: https://svn.code.sf.net/p/xcat/code/xcat-core/trunk@2559 8638fb3e-16cb-4fca-ae20-7b5d299a9bcd
This commit is contained in:
mellor 2009-01-05 18:02:00 +00:00
parent 42908ab23b
commit 17b94163f0
2 changed files with 26 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
xCAT-server/lib/xcat/plugins/rollupdate.pm
View File

@ -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";

6
xCAT-server/share/xcat/rollupdate/ll.tmpl
View File

@ -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