#!/bin/sh
#
#NADDS Agent - http://nadds.drmadcow.net/
#disk i/o load agent
#Agent Author: David Blizard
#Date: 2003-May-26 (V3.0)

. /usr/local/bcnu/etc/bcnuenv
. $BCNUHOME/agent/agent_head

#setup file handles and variables
ERRFILE=$BCNUTMP/${BCNUAGENT}.err
OUTFILE=$BCNUTMP/${BCNUAGENT}.out
TMPFILE=$BCNUTMP/${BCNUAGENT}.tmp
MSGFILE=$BCNUTMP/${BCNUAGENT}.msg
STATFILE=$BCNUTMP/${BCNUAGENT}.status

#dont touch these variables
ERROR=0
WARN=0
ITERATOR=0
TPS_ERR_COUNT=0
BRD_ERR_COUNT=0
BWT_ERR_COUNT=0

#remove any old files
rm -f $ERRFILE $OUTFILE $MSGFILE $TMPFILE $STATFILE 2>/dev/null

#initialize status file
echo "0" > $STATFILE 2>/dev/null

#get and set BCNU paramaters
param=$BCNUPARAM
bcnu_param

MAX_TPS=${p1:=1000}
MAX_BLKRDPS=${p2:=1000}
MAX_BLKWTPS=${p3:=1000}
INTERVAL=${p4:=1}
COUNT=${p5:=10}
MAX_WARN=${p6:=1}

#compensate for throwing out first line of iostat command
MAXCOUNT=$COUNT
COUNT=$[ $COUNT + 1 ]

#get timestamp
DATE=`date`
#run the iostat command 
iostat -d $INTERVAL $COUNT | sed '4,/^Device/d' | sed '1,2d' | sed '/^$/,/^Device/d' > $OUTFILE 2>/dev/null
if [ $? -ne 0 ] ; then
	BCNUMSG="$BCNU -m 'error - iostat command failed' -f $ERRFILE -e $BCNU_ERR -t $BCNUHOSTTYPE $BCNUHOST"
	bcnu_err_send
	exit 1
fi

#format the data
sed '1d' $OUTFILE | sort > $TMPFILE 2>/dev/null

#prepare message file          	
echo -e "\n|====Device Status====|\n" >> $MSGFILE 2>/dev/null

cat $TMPFILE |
while read DEVICE TPS BLK_READ_PS BLK_WRTN_PS BLK_READ BLK_WRITTEN  2>/dev/null
do
	#chop off decimals, sorry no rounding
	TPS=`echo $TPS | sed 's/\.[0-9]*//'`
	BLK_READ_PS=`echo $BLK_READ_PS | sed 's/\.[0-9]*//'`
	BLK_WRTN_PS=`echo $BLK_WRTN_PS | sed 's/\.[0-9]*//'` 		
	
	#check for tps error and set status flag		
	if [ $TPS -gt $MAX_TPS ] ;then
		TPS_ERR_COUNT=$[ $TPS_ERR_COUNT + 1 ]
		if [ $TPS_ERR_COUNT -gt $MAX_WARN ] ;then
			ERROR=1					
		else
			WARN=1
		fi		
	fi
		
	#check for blocks read error and set status flag
	if [ $BLK_READ_PS -gt $MAX_BLKRDPS ] ;then
		BRD_ERR_COUNT=$[ $BRD_ERR_COUNT + 1 ]
		if [ $BRD_ERR_COUNT -gt $MAX_WARN ] ;then
			ERROR=1
		else
			WARN=1
		fi		
	fi
	
	#check for blocks written error and set status flag
	if [ $BLK_WRTN_PS -gt $MAX_BLKWTPS ] ;then
		BWT_ERR_COUNT=$[ $BWT_ERR_COUNT + 1 ]
		if [ $BWT_ERR_COUNT -gt $MAX_WARN ] ;then
			ERROR=1
		else
			WARN=1
		fi	
	fi
	
	#count what report we are on
	ITERATOR=$[ $ITERATOR + 1 ]

	#if on last report log device status
	if [ $ITERATOR -eq $MAXCOUNT ] ;then
		if [ $ERROR -eq 1 ] ;then
			#set error level to status file
			echo "1" > $STATFILE 2>/dev/null
			echo -e "Device: $DEVICE Status: ERROR\n\t$TPS_ERR_COUNT Transfers Per Second errors\n\t$BRD_ERR_COUNT Blocks Read Per Second errors\n\t$BWT_ERR_COUNT Blocks Written Per Second errors" >> $MSGFILE 2>/dev/null
		elif [ $WARN -eq 1 ] ;then
			#read status level set to warn if ok 		
			read STATUS < $STATFILE 2>/dev/null
			if [ $STATUS -eq 0 ] ;then
				echo "2" > $STATFILE 2>/dev/null
			fi
			echo -e "Device: $DEVICE Status: WARNING\n\t$TPS_ERR_COUNT Transfers Per Second errors\n\t$BRD_ERR_COUNT Blocks Read Per Second errors\n\t$BWT_ERR_COUNT Blocks Written Per Second errors" >> $MSGFILE 2>/dev/null	
		else
			echo -e "Device: $DEVICE Status: OK\n\t$TPS_ERR_COUNT Transfers Per Second errors\n\t$BRD_ERR_COUNT Blocks Read Per Second errors\n\t$BWT_ERR_COUNT Blocks Written Per Second errors" >> $MSGFILE 2>/dev/null
		fi
		#initialize for next device
		ITERATOR=0
		ERROR=0
		WARN=0
	fi
done
	#send bcnu message and data
	echo -e "\n|====iostat output - $DATE====|\n" >> $MSGFILE 2>/dev/null
	cat $OUTFILE >> $MSGFILE 2>/dev/null
	read STATUS < $STATFILE 2> /dev/null
	if [ $STATUS -eq 1 ] ;then
		BCNUMSG="$BCNU -m 'error -  Disk Load Error Alert' -f $MSGFILE -e $BCNU_ERR -t $BCNUHOSTTYPE $BCNUHOST"
		bcnu_err_send
	elif [ $STATUS -eq 2 ] ;then
		BCNUMSG="$BCNU -m 'warning - Disk Load Warning Alert' -f $MSGFILE -e $BCNU_WARN -t $BCNUHOSTTYPE $BCNUHOST"
		bcnu_warn_send
	else 
		BCNUMSG="$BCNU -m 'ok - Disks OK' -f $MSGFILE -e $BCNU_OK -t $BCNUHOSTTYPE $BCNUHOST"
		bcnu_send
	fi