added support to install predefined sensors to nodes for RMC monitoring
git-svn-id: https://svn.code.sf.net/p/xcat/code/xcat-core/trunk@3564 8638fb3e-16cb-4fca-ae20-7b5d299a9bcd
This commit is contained in:
linggao
parent
4c1df7b8a2
commit
be4d2366b6
@ -1,7 +1,15 @@
|
||||
#!/usr/bin/perl
|
||||
|
||||
my $cmd;
|
||||
if (-r "/var/xcat/rmcmon/scripts/monerrorlog") {
|
||||
$cmd="/var/xcat/rmcmon/scripts/monerrorlog";
|
||||
} else {
|
||||
$cmd="/opt/xcat/sbin/rmcmon/monerrorlog";
|
||||
}
|
||||
|
||||
$RES::Sensor{'ErrorLogSensor'} = {
|
||||
Name => q(ErrorLogSensor),
|
||||
Command => "/etc/xcat/rmcmon/scripts/monerrorlog",
|
||||
Command => "$cmd",
|
||||
UserName => q(root),
|
||||
RefreshInterval => q(60),
|
||||
ControlFlags => q(4),
|
||||
|
||||
@ -65,137 +65,147 @@ then
|
||||
exit 1
|
||||
fi
|
||||
|
||||
|
||||
|
||||
#check if the resource is defined
|
||||
result=`/usr/bin/lsrsrc-api -s IBM.MCP::"NodeID=0x$MS_NODEID"::MNName 2>&1`
|
||||
if [ $? -gt 0 ]; then
|
||||
# echo $result
|
||||
p="2612-023" #resource not found
|
||||
result1=`awk -v a="$result" -v b="$p" 'BEGIN{print index(a,b)}'`
|
||||
if [ $result1 -gt 0 ]; then
|
||||
if [ $ADD -eq -1 ]; then
|
||||
exit 0; #resource already removed, do nothing
|
||||
fi
|
||||
else
|
||||
echo "$result"
|
||||
logger xCAT "$result"
|
||||
exit 1
|
||||
fi
|
||||
else
|
||||
if [ $ADD -eq 1 ]; then
|
||||
exit 0; #resource already defined, do nothing
|
||||
fi
|
||||
fi
|
||||
|
||||
#ask RMC take the new nodeid
|
||||
if [[ $NODEID != "" ]]; then
|
||||
oldid=`/usr/sbin/rsct/bin/lsnodeid`
|
||||
oldidU=`echo "$oldid" | awk '{ print toupper($1) }'`
|
||||
newidU=`echo "$NODEID" | awk '{ print toupper($1) }'`
|
||||
if [ $oldidU != $newidU ]; then
|
||||
if [ -e /usr/sbin/rsct/install/bin/recfgctnid ]; then
|
||||
result=`/usr/sbin/rsct/install/bin/recfgctnid $NODEID 2>&1`
|
||||
if [ $? -ne 0 ]; then
|
||||
echo "Reset nodeid: $result"
|
||||
logger xCAT "Reset nodeid: $result"
|
||||
fi
|
||||
else
|
||||
#stop all rmc daemons
|
||||
/usr/sbin/rsct/bin/rmcctrl -z
|
||||
|
||||
#get new nodeid into id files
|
||||
if [ -e /var/ct/cfg/ct_node_id ]; then
|
||||
sed s/^[^\#].*$/$NODEID/ /var/ct/cfg/ct_node_id > /tmp/id.tmp
|
||||
else
|
||||
if [ -e /etc/ct_node_id ]; then
|
||||
sed s/^[^\#].*$/$NODEID/ /etc/ct_node_id > /tmp/id.tmp
|
||||
if [ $ADD -eq 1 ]; then
|
||||
#check if the resource is defined
|
||||
result=`/usr/bin/lsrsrc-api -s IBM.MCP::"NodeID=0x$MS_NODEID"::MNName 2>&1`
|
||||
if [ $? -gt 0 ]; then #node defined
|
||||
#ask RMC take the new nodeid
|
||||
if [[ $NODEID != "" ]]; then
|
||||
oldid=`/usr/sbin/rsct/bin/lsnodeid`
|
||||
oldidU=`echo "$oldid" | awk '{ print toupper($1) }'`
|
||||
newidU=`echo "$NODEID" | awk '{ print toupper($1) }'`
|
||||
if [ $oldidU != $newidU ]; then
|
||||
if [ -e /usr/sbin/rsct/install/bin/recfgctnid ]; then
|
||||
result=`/usr/sbin/rsct/install/bin/recfgctnid $NODEID 2>&1`
|
||||
if [ $? -ne 0 ]; then
|
||||
echo "Reset nodeid: $result"
|
||||
logger xCAT "Reset nodeid: $result"
|
||||
fi
|
||||
else
|
||||
echo $NODEID > /tmp/id.tmp
|
||||
#stop all rmc daemons
|
||||
/usr/sbin/rsct/bin/rmcctrl -z
|
||||
|
||||
#get new nodeid into id files
|
||||
if [ -e /var/ct/cfg/ct_node_id ]; then
|
||||
sed s/^[^\#].*$/$NODEID/ /var/ct/cfg/ct_node_id > /tmp/id.tmp
|
||||
else
|
||||
if [ -e /etc/ct_node_id ]; then
|
||||
sed s/^[^\#].*$/$NODEID/ /etc/ct_node_id > /tmp/id.tmp
|
||||
else
|
||||
echo $NODEID > /tmp/id.tmp
|
||||
fi
|
||||
fi
|
||||
cp /tmp/id.tmp /var/ct/cfg/ct_node_id
|
||||
cp /tmp/id.tmp /etc/ct_node_id
|
||||
rm /tmp/id.tmp
|
||||
|
||||
#reconfig RMC
|
||||
result=`/usr/sbin/rsct/install/bin/recfgct -s 2>&1`
|
||||
if [ $? -ne 0 ]; then
|
||||
echo "RMC setup on node $NODE: Cannot reconfig RSCT with new node id"
|
||||
logger xCAT "RMC setup on node $NODE: Cannot reconfig RSCT with new node id"
|
||||
fi
|
||||
fi
|
||||
fi
|
||||
cp /tmp/id.tmp /var/ct/cfg/ct_node_id
|
||||
cp /tmp/id.tmp /etc/ct_node_id
|
||||
rm /tmp/id.tmp
|
||||
|
||||
#reconfig RMC
|
||||
result=`/usr/sbin/rsct/install/bin/recfgct -s 2>&1`
|
||||
if [ $? -ne 0 ]; then
|
||||
echo "RMC setup on node $NODE: Cannot reconfig RSCT with new node id"
|
||||
logger xCAT "RMC setup on node $NODE: Cannot reconfig RSCT with new node id"
|
||||
fi
|
||||
|
||||
PID=`/bin/ps -ef | /bin/grep rmcd | /bin/grep -v grep | /bin/awk '{print $2}'`
|
||||
if [ -z "$PID" ]; then
|
||||
#restart rmc daemon
|
||||
result=`/usr/sbin/rsct/bin/rmcctrl -s 2>&1`;
|
||||
if [ $? -gt 0 ]; then
|
||||
echo "RMC deamon cannot be started on node $NODE:$result"
|
||||
logger xCAT "RMC deamon cannot be started on node $NODE:$result"
|
||||
exit 1;
|
||||
fi
|
||||
fi
|
||||
fi
|
||||
fi
|
||||
|
||||
#wait for RMC daemon to fully functional
|
||||
RETRY_LIMIT=20 # number of 2 second retry intervals (40 second total)
|
||||
let i=$RETRY_LIMIT
|
||||
while :
|
||||
do
|
||||
lsrsrc > /dev/null 2>&1
|
||||
rc=$?
|
||||
if [[ $rc -eq 0 ]]
|
||||
then
|
||||
break
|
||||
fi
|
||||
|
||||
i=$((i - 1))
|
||||
if (( i > 0 ))
|
||||
then
|
||||
sleep 2
|
||||
else
|
||||
break
|
||||
fi
|
||||
done
|
||||
if [[ $rc -ne 0 ]]
|
||||
then
|
||||
logger xCAT "RMC subsystem has not started"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
#enable remote client connection
|
||||
/usr/sbin/rsct/bin/rmcctrl -p; /usr/bin/refrsrc IBM.MCP
|
||||
|
||||
#get IP address of MS
|
||||
ms_ip=$MONMASTER
|
||||
result=`ping -c1 $MONMASTER 2>&1`
|
||||
if [ $? -eq 0 ]; then
|
||||
index1=`expr index "$result" "\("`
|
||||
index2=`expr index "$result" "\)"`
|
||||
#ms_ip=${result:$index1+1:$index2-$index1-2}
|
||||
pos=`expr $index1 + 1`
|
||||
length=`expr $index2 - $index1`
|
||||
length=`expr $length - 1`
|
||||
ms_ip=`expr substr "$result" $pos $length`
|
||||
else
|
||||
echo "RMC setup on node $NODE:$result"
|
||||
logger xCAT "RMC setup on node $NODE:$result"
|
||||
fi
|
||||
|
||||
PID=`/bin/ps -ef | /bin/grep rmcd | /bin/grep -v grep | /bin/awk '{print $2}'`
|
||||
if [ -z "$PID" ]; then
|
||||
#restart rmc daemon
|
||||
result=`/usr/sbin/rsct/bin/rmcctrl -s 2>&1`;
|
||||
if [ $? -gt 0 ]; then
|
||||
echo "RMC deamon cannot be started on node $NODE:$result"
|
||||
logger xCAT "RMC deamon cannot be started on node $NODE:$result"
|
||||
exit 1;
|
||||
fi
|
||||
fi
|
||||
|
||||
#wait for RMC daemon to fully functional
|
||||
RETRY_LIMIT=20 # number of 2 second retry intervals (40 second total)
|
||||
let i=$RETRY_LIMIT
|
||||
while :
|
||||
do
|
||||
lsrsrc > /dev/null 2>&1
|
||||
rc=$?
|
||||
if [[ $rc -eq 0 ]]
|
||||
then
|
||||
break
|
||||
#define resource in IBM.MCP class on node
|
||||
result1=`/usr/bin/mkrsrc-api IBM.MCP::MNName::"$NODE"::KeyToken::"$MONMASTER"::IPAddresses::"{\"$ms_ip\"}"::NodeID::0x$MS_NODEID 2>&1`
|
||||
if [ $? -gt 0 ]; then
|
||||
echo "Define resource in IBM.MCP class on node $NODE. result=$result1"
|
||||
logger xCAT "Define resource in IBM.MCP class on node $NODE. result=$result1"
|
||||
exit 1
|
||||
fi
|
||||
fi
|
||||
|
||||
i=$((i - 1))
|
||||
if (( i > 0 ))
|
||||
then
|
||||
sleep 2
|
||||
else
|
||||
break
|
||||
#create predefined sensors
|
||||
if [[ -z "$NFSSERVER" ]]; then
|
||||
NFSSERVER=$MASTER
|
||||
fi
|
||||
done
|
||||
if [[ $rc -ne 0 ]]
|
||||
then
|
||||
logger xCAT "RMC subsystem has not started"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
|
||||
if [ $ADD -eq 1 ]; then
|
||||
#enable remote client connection
|
||||
/usr/sbin/rsct/bin/rmcctrl -p; /usr/bin/refrsrc IBM.MCP
|
||||
|
||||
#get IP address of MS
|
||||
ms_ip=$MONMASTER
|
||||
result=`ping -c1 $MONMASTER 2>&1`
|
||||
if [ $? -eq 0 ]; then
|
||||
index1=`expr index "$result" "\("`
|
||||
index2=`expr index "$result" "\)"`
|
||||
#ms_ip=${result:$index1+1:$index2-$index1-2}
|
||||
pos=`expr $index1 + 1`
|
||||
length=`expr $index2 - $index1`
|
||||
length=`expr $length - 1`
|
||||
ms_ip=`expr substr "$result" $pos $length`
|
||||
else
|
||||
echo "RMC setup on node $NODE:$result"
|
||||
logger xCAT "RMC setup on node $NODE:$result"
|
||||
fi
|
||||
|
||||
#define resource in IBM.MCP class on node
|
||||
result1=`/usr/bin/mkrsrc-api IBM.MCP::MNName::"$NODE"::KeyToken::"$MONMASTER"::IPAddresses::"{\"$ms_ip\"}"::NodeID::0x$MS_NODEID 2>&1`
|
||||
if [ $? -gt 0 ]; then
|
||||
echo "Define resource in IBM.MCP class on node $NODE. result=$result1"
|
||||
logger xCAT "Define resource in IBM.MCP class on node $NODE. result=$result1"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
#TODO: create predefined sensors
|
||||
mkdir -p /var/xcat/rmcmon
|
||||
rm -f -R /var/xcat/rmcmon
|
||||
mkdir -p /tmp/postage/rmcmon
|
||||
rm -f -R /tmp/postage/rmcmon/*
|
||||
cd /tmp/postage/rmcmon
|
||||
wget -l inf -N -r --waitretry=10 --random-wait --retry-connrefused -t 0 -T 60 ftp://$NFSSERVER/post/xcat/rmcmon 2> /tmp/wget.log
|
||||
mv $NFSSERVER/post/xcat/rmcmon /var/xcat/rmcmon;
|
||||
rm -rf $NFSSERVER
|
||||
chmod 755 /var/xcat/rmcmon/scripts/*
|
||||
/var/xcat/rmcmon/scripts/mkrmcresources /var/xcat/rmcmon/resources
|
||||
else
|
||||
#check if the resource is defined
|
||||
result=`/usr/bin/lsrsrc-api -s IBM.MCP::"NodeID=0x$MS_NODEID"::MNName 2>&1`
|
||||
if [ $? -gt 0 ]; then
|
||||
# echo $result
|
||||
p="2612-023" #resource not found
|
||||
result1=`awk -v a="$result" -v b="$p" 'BEGIN{print index(a,b)}'`
|
||||
if [ $result1 -gt 0 ]; then
|
||||
if [ $ADD -eq -1 ]; then
|
||||
exit 0; #resource already removed, do nothing
|
||||
fi
|
||||
else
|
||||
echo "$result"
|
||||
logger xCAT "$result"
|
||||
exit 1
|
||||
fi
|
||||
fi
|
||||
|
||||
#remove resource in IBM.MCP class on the node
|
||||
result2=`/usr/bin/rmrsrc-api -s IBM.MCP::"NodeID=0x$MS_NODEID" 2>&1`
|
||||
if [ $? -gt 0 ]; then
|
||||
@ -204,6 +214,7 @@ else
|
||||
exit 1
|
||||
fi
|
||||
fi
|
||||
|
||||
echo "done"
|
||||
exit 0;
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user