xCAT/xcat-core
2
0
Fork 0
mirror of https://github.com/xcat2/xcat-core.git synced 2025-10-31 03:12:30 +00:00
Code Issues Releases Wiki Activity

some responses for monitoring batch events on SN or HMC

Browse Source 
git-svn-id: https://svn.code.sf.net/p/xcat/code/xcat-core/trunk@8111 8638fb3e-16cb-4fca-ae20-7b5d299a9bcd
This commit is contained in:
linggao
2010-11-09 21:13:54 +00:00
parent 69527bc067
commit d19917faf4
4 changed files with 185 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
xCAT-rmc/resources/mn/IBM.EventResponse/MsgRootBatchEvents_H.pm Normal file
View File

@@ -0,0 +1,8 @@
#!/usr/bin/perl
$RES::EventResponse{'MsgRootBatchEvents_H'} = {
Name => q(MsgRootBatchEvents_H),
Locked => q(0),
Actions => q({[msgEvent,{127},{0},{86400},/opt/xcat/sbin/rmcmon/msg-hierarchical-batch-event root -d,3,0,0,0,{},0]}),
};
1;

19
xCAT-rmc/scripts/email-hierarchical-batch-event
View File

@@ -85,7 +85,16 @@ my $dirname="/tmp/batch_process/";
if (! -d $dirname) {
mkdir($dirname);
}
my $cmd="scp $sn_name:$filename $dirname/$bn";
my $cmd;
my $isHMC=0;
if ($filename =~ /\/home\/hscroot\/tmp/) {
$isHMC=1;
$cmd="scp hscroot\@$sn_name:$filename $dirname/$bn";
}
else {
$cmd="scp $sn_name:$filename $dirname/$bn";
}
my $rc=`$cmd 2>&1`;
if ($? != 0) {
$msg .= "$rc\n";
@@ -126,14 +135,18 @@ for ($count = 1; $count <= $num_events; $count++) {
}
#send the mail out
`echo "$msg"| mail -s "$severity $type: $ENV{ERRM_COND_NAME}" $user`;
`echo "$msg"| mail -s "xxx$severity $type: $ENV{ERRM_COND_NAME}" $user`;
#delete the batch file on the mn
`rm $dirname/$bn`;
#remove the batch file on the sn if needed
if ($delete) {
`ssh -l root $sn_name "rm $filename"`;
if ($isHMC) {
`ssh -l hscroot $sn_name "rm $filename"`;
} else {
`ssh -l root $sn_name "rm $filename"`;
}
}
exit;

159
xCAT-rmc/scripts/msg-hierarchical-batch-event Executable file
View File

@@ -0,0 +1,159 @@
#!/usr/bin/env perl
# IBM(c) 2007 EPL license http://www.eclipse.org/legal/epl-v10.html
# Script to demonstrate how to parse out the event info that comes from a
# Condition that is watching another Condition in a hierarchical cluster environment.
# The condition that is being watched is a batch event. This script will go th the
# node that has the watched condition, get the event batch file. And parse the file
# and then send email to a user with the detailed info of the events in the batch file.
# The batch file will be then deleted if -d flag is set.
# To use this script, create a Response that invokes this script with user id as the input.
# Then associate the Response with a Condition that is watching Conditions on other
# nodes. When the Conditions occur, this script will wall the relevant event information.
# This info that is passed into this sample is structured as follows:
# EMS Condition/Event:
# $ENV{ERRM_COND_NAME}
# $ENV{ERRM_TYPEID}
# ...
# $ENV{ERRM_NODE_NAME} --> SN
# $ENV{ERRM_RSRC_NAME} --> SN Condition
# $ENV{ERRM_VALUE} --> SN Event: LastBatchedEventFile
# $Saved
# $FileName
# $Size
use strict;
use Getopt::Std;
use POSIX qw(strftime);
use File::Basename;
use File::Path;
my $where = join(' ', @ARGV);
my $delete=0;
if ($where =~ /-d/) {
$delete=1;
}
$where =~ s/-d//; #remove -d
# Convert Severity and Type environment variables from integers to strings
my @COND_SEVERITY = qw(Informational Warning Critical);
my @TYPE = ('event', 'rearm event');
my $severity=$COND_SEVERITY[$ENV{ERRM_COND_SEVERITYID}];
my $type=$TYPE[$ENV{ERRM_TYPEID }];
# Parse the ERRM_VALUE attribute, which will contain the EventOccurred structured data variable from the Condition class
# The fields in this structured data variable are documented below where we parse them out.
my $event = $ENV{ERRM_VALUE};
$event =~ s/^\[(.*)\]$/$1/; # SD variables have square brackets around them
# This parse the LastEvent
my ( # split the SD into the following fields:
$Saved, # batch file has been saved
$filename, # location of the batch file
$Size, # The size of the batch file
) = split(/,/, $event);
my $sn_condname=$ENV{ERRM_RSRC_NAME};
my $sn_name=$ENV{ERRM_NODE_NAME};
my (
$EventTime, # Time of event expressed in seconds since 1/1/1970
$EventTimeMicros, # Number of microseconds past EventTime
) = split(/,/, $ENV{ERRM_TIME});
my $msg;
$msg .= "The following $type occurred:\n";
$msg .= " Event Time: " . convertTime($EventTime) . "\n";
$msg .= " Condition name: $ENV{ERRM_COND_NAME}\n";
$msg .= " Severiry: $severity\n";
$msg .= " Condition being monitored: $sn_condname\n";
$msg .= " Node where the condition was monitored: $sn_name\n"; # do we have to use $ERRM_NODE_NAMELIST here?
$msg .= " SN batch files name: $filename\n";
#copy the batch file from sn to mn
$filename =~ s/"//g;
my $bn=basename($filename);
#printf stderr "ful path :$filename\n";
#printf stderr "base name:$bn\n";
my $dirname="/tmp/batch_process/";
if (! -d $dirname) {
mkdir($dirname);
}
my $cmd;
my $isHMC=0;
if ($filename =~ /\/home\/hscroot\/tmp/) {
$isHMC=1;
$cmd="scp hscroot\@$sn_name:$filename $dirname/$bn";
}
else {
$cmd="scp $sn_name:$filename $dirname/$bn";
}
my $rc=`$cmd 2>&1`;
if ($? != 0) {
$msg .= "$rc\n";
}
#now process the batch file
open(FILE1, "<$dirname/$bn");
readline(FILE1);#skip first 2 lines
readline(FILE1);
my $line1=readline(FILE1);
my @aTemp=split(/=/, $line1);
my $num_events=$aTemp[1];
close(FILE1);
my $count;
for ($count = 1; $count <= $num_events; $count++) {
my $content=`sed -n "/Event $count:/,/ERRM_COND_BATCH/ p" $dirname/$bn`;
my @content_array=split(/\n/, $content);
pop(@content_array); #get rid of last line
shift(@content_array); #get rid of firt line
my %content_hash=();
foreach(@content_array) {
/([^\=]+)\=(.*)/;
$content_hash{$1}=$2;
}
my $event = "Event count: $count\n";
#$event .= " Event time: " . $content_hash{ERRM_TIME} . "\n";
$event .= " Node where the event occurred: " . $content_hash{ERRM_NODE_NAMELIST} . "\n";
$event .= " Resource class: " . $content_hash{ERRM_RSRC_CLASS_PNAME} . "\n";
$event .= " Resource name: " . $content_hash{ERRM_RSRC_NAME} . "\n";
$event .= " Attribute name: " . $content_hash{ERRM_ATTR_PNAME} . "\n";
$event .= " Attribute value: " . $content_hash{ERRM_VALUE} . "\n\n";
$msg .= $event;
}
#send the mail out
my $rc = system(qq(echo "$msg" | write $where)) >> 8;
#delete the batch file on the mn
`rm $dirname/$bn`;
#remove the batch file on the sn if needed
if ($delete) {
if ($isHMC) {
`ssh -l hscroot $sn_name "rm $filename"`;
} else {
`ssh -l root $sn_name "rm $filename"`;
}
}
exit $rc;
# convert time string
sub convertTime {
my ($seconds, $micro) = split(/\./, $_[0]);
return strftime("%A %D %T", localtime($seconds));
}

4
xCAT-rmc/scripts/msg-hierarchical-event
View File

@@ -4,7 +4,7 @@
# Script to demonstrate how to parse out the event info that comes from a
# Condition that is watching another Condition in a hierarchical cluster environment.
# To use this script, create a Response that invokes this script with email address as the input.
# To use this script, create a Response that invokes this script with user id as the input.
# Then associate the Response with a Condition that is watching Conditions on other
# nodes. When the Conditions occur, this script will wall the relevant event information.
@@ -79,7 +79,7 @@ $msg .= " MN Condition: $ENV{ERRM_COND_NAME}\n";
$msg .= " SN Condition: $ENV{ERRM_RSRC_NAME}\n";
$msg .= " SN: $ENV{ERRM_NODE_NAME}\n"; # do we have to use $ERRM_NODE_NAMELIST here?
$msg .= " Node: $NodeName\n";
$msg .= " Resource: $ResourceName\n";
$msg .= " Resource Name: $ResourceName\n";
$msg .= " Time: " . convertTime($EventTime) . " \n";
if (length($valuesMsg)) {
$msg .= " Attributes:\n";