#!/bin/sh # # Sample postscript to check appstatus for GPFS,LAPI and LoadL on Linux system. # Report the appstatus to xcatd on MN. # Set the default timeout to 5 mins. TIMEOUT=300 INTERVAL=15 PASSED=0 # The number of items to check, currently including: # gpfs-daemon, gpfs-quorum, gpfs-filesystem, lapi-pnsd, loadl-schedd, loadl-startd ITEMNUM=6 gpfsd="hpcbootstatus gpfs-daemon=up" gpfsq="hpcbootstatus gpfs-quorum=achieved" # Get server IP from xcatinfo file if [ -f /opt/xcat/xcatinfo ]; then SIP=`cut -d= -f2 /opt/xcat/xcatinfo` if [ -n "$SIP" ]; then MASTER=$SIP fi else echo "xCAT management server IP can't be determined.\nexiting..." exit -1 fi # add HPCbootstatus.linuxinit to /etc/init.d for diskful node reboot cat /xcatpost/HPCbootstatus.linuxinit >/etc/init.d/HPCbootstatus.linuxinit # Change permissions chmod 755 /etc/init.d/HPCbootstatus.linuxinit chkconfig --add HPCbootstatus.linuxinit # Use a loop while [ true ]; do COUNT=0 # Query gpfs-daemon,mmfsd # Check 1191 port port=`netstat -an|grep 1191` if [ "$port" != "" ]; then # Sent message to xcatd updateflag.awk $MASTER 3002 "$gpfsd" echo "HPCbootstatus.linux: Reporting $gpfsd to xcatd on $MASTER!" let COUNT=$COUNT+1 fi # Query gpfs-quorum quorum=`/usr/lpp/mmfs/bin/mmgetstate -s|grep achieved` if [ "$quorum" != "" ]; then # Sent message to xcatd updateflag.awk $MASTER 3002 "$gpfsq" echo "HPCbootstatus.linux: Reporting $gpfsq to xcatd on $MASTER!" let COUNT=$COUNT+1 fi # Query gpfs-filesystem fs=`/bin/mount |grep "type gpfs"|awk '{print $3}'|tr '\n' '!'` if [ "$fs" != "" ]; then # Sent message to xcatd updateflag.awk $MASTER 3002 "hpcbootstatus gpfs-filesystem=$fs" echo "HPCbootstatus.linux: Reporting gpfs-filesystem=$fs to xcatd on $MASTER!" let COUNT=$COUNT+1 fi # Query lapi-pnsd lapi=`/usr/bin/lssrc -s pnsd |/usr/bin/awk '(NR==2){print $3}'` if [ "$lapi" != "" ]; then # Sent message to xcatd updateflag.awk $MASTER 3002 "hpcbootstatus lapi-pnsd=$lapi" echo "HPCbootstatus.linux: Reporting lapi-pnsd=$lapi to xcatd on $MASTER!" let COUNT=$COUNT+1 fi # Query loadl-schedd hn=`hostname| tr '[A-Z]' '[a-z]'` llsc=`/usr/bin/llrstatus -h $hn -r %sca %scs` if [ "$llsc" != "" ]; then # Sent message to xcatd updateflag.awk $MASTER 3002 "hpcbootstatus loadl-schedd=$llsc" echo "HPCbootstatus.linux: Reporting loadl-schedd=$llsc to xcatd on $MASTER!" let COUNT=$COUNT+1 fi # Query loadl-startd llst=`/usr/bin/llrstatus -h $hn -r %sta %sts` if [ "$llst" != "" ]; then # Sent message to xcatd updateflag.awk $MASTER 3002 "hpcbootstatus loadl-startd=$llst" echo "HPCbootstatus.linux: Reporting loadl-startd=$llst to xcatd on $MASTER!" let COUNT=$COUNT+1 fi if [ $COUNT = $ITEMNUM ]; then exit 0 fi if [ $PASSED -gt $TIMEOUT ]; then exit 1 fi let PASSED=$PASSED+$INTERVAL sleep $INTERVAL echo "HPCbootstatus.linux: passed $PASSED" done exit 0