hw-test/scripts/wrapper-fio

#!/bin/bash

source test-variables

DISK=$1
if [[ -z $DISK ]] ; then
	echo "Missing DISK parameter"
	exit 11
fi

SERIAL=$2
if [[ -z $SERIAL ]] ; then
	echo "Missing SERIAL parameter"
	exit 12
fi

MY_PROC=$3
if [[ -z $MY_PROC ]] ; then
	echo "Missing parameter 2: MY_PROC"
	exit 16
fi

if [[ ! -b /dev/$DISK ]] ; then
	echo "Device /dev/$DISK is not a block special file"
	exit 13
fi

DISK_SERIAL=`smartctl -i /dev/$DISK \
	| grep -i 'serial number' \
	| cut -d ':' -f 2 \
	| sed 's/\s\+//'`

if [[ -z $DISK_SERIAL ]] ; then
	echo "Empty serial number in disk /dev/$DISK"
	exit 14
fi

if [[ $DISK_SERIAL != $SERIAL ]] ; then
	echo "Serial numbers $SERIAL and $DISK_SERIAL for $DISK do not match"
	exit 15
fi

# TODO: Debug limit. Comment out for production
#LST_BLCK=10000000
SMARTLOG=300 # check every 5 minutes
#SMARTLOG=60 # check every minute
# TODO: Debug variable. Comment out for production
#SMARTLOG=15 # Check every second
BLK_SIZE=`cat /sys/block/$DISK/queue/physical_block_size`

cat > $OUTPUT_DIR/disk/$DISK.job << EOF
[global]
ioengine=libaio
direct=1
gtod_reduce=1
bs=$BLK_SIZE
iodepth=64
# TODO: Uncomment for production
size=4G
# TODO: Debug variable. Comment out for production
#size=16M
readwrite=randrw
rwmixread=75
directory=/mnt/$DISK

[file1]
name=test
filename=test

EOF

BEGIN=`date +%s`
# Check SMART overall self test
ERROR=`smartctl -H /dev/$DISK \
	| grep '\(overall-health\|Health Status\)' \
	| grep -v '\(PASSED\|OK\)'
`
if [[ -n $ERROR ]] ; then
	smartctl -H /dev/$DISK \
		> $OUTPUT_DIR/disk/$SERIAL.smart.error
else
	ERROR=`smartctl -A -f brief /dev/$DISK \
		| parse-smart-attr \
		| grep ERROR \
		| cut -d ':' -f 2
	`
	# TODO: Debug variable. Comment out for production
	ERROR=
	if [[ -n $ERROR ]] ; then
		RAW_ERROR=$ERROR
		ERROR=''
		for TOKEN in $RAW_ERROR ; do
			ERROR="$ERROR $TOKEN"
		done
		smartctl -A -f brief /dev/$DISK \
		| parse-smart-attr \
		| sed "s/ERROR/$RES_ERROR/g" \
			> $OUTPUT_DIR/disk/$SERIAL.smart.error
	fi
fi
if [[ -z $ERROR ]] ; then
	# Reset partition table to zero
	dd if=/dev/zero of=/dev/${DISK} bs=$BLK_SIZE count=128 \
		>> /run/$SERIAL.fio.out 2>&1
	parted /dev/${DISK} mklabel gpt mkpart primary 1 128G \
		>> /run/$SERIAL.fio.out 2>&1
	sleep 1
	if [[ ! -b /dev/${DISK}1 ]] ; then
		echo "Device /dev/${DISK}1 is not a block special file"
		exit 17
	fi
	mkfs.ext4 /dev/${DISK}1 \
		>> /run/$SERIAL.fio.out 2>&1
	mkdir -pv /mnt/${DISK}
	mount /dev/${DISK}1 /mnt/${DISK}
	MOUNTED=`cat /proc/mounts | grep ${DISK}1`
	if [[ -z $MOUNTED ]] ; then
		echo "Device /dev/${DISK}1 is not mounted"
		exit 18
	fi
	fio --output $OUTPUT_DIR/disk/$SERIAL.fio.out \
		$OUTPUT_DIR/disk/$DISK.job  &
	TEST_PID=`jobs -p %+`
	echo $TEST_PID > /run/$SERIAL.fio.pid
	while [[ -d /proc/$TEST_PID ]] ; do
		# Check SMART error log
		ERROR=`smartctl -l error /dev/$DISK \
			| parse-smart-error \
			| sed 's/://g'
		`
		# TODO: Debug variable. Comment out for production
		ERROR=
		if [[ -n $ERROR ]] ; then
			smartctl -l error /dev/$DISK \
				> $OUTPUT_DIR/disk/$SERIAL.smart.error
			kill $TEST_PID
			sleep 5
		fi
		sleep $SMARTLOG
	done
	umount -f /mnt/${DISK}
	# Reset partition table to zero
	dd if=/dev/zero of=/dev/${DISK} bs=$BLK_SIZE count=128 \
		>> /run/$SERIAL.fio.out 2>&1
	if [[ -z $ERROR ]] ; then
		# Check smart attributes after the fio test
		ERROR=`smartctl -A -f brief /dev/$DISK \
			| parse-smart-attr \
			| grep ERROR \
			| cut -d ':' -f 2
		`
		# TODO: Debug variable. Comment out for production
		ERROR=
		if [[ -n $ERROR ]] ; then
			RAW_ERROR=$ERROR
			ERROR=''
			for TOKEN in $RAW_ERROR ; do
				ERROR="$ERROR $TOKEN"
			done
			smartctl -A -f brief /dev/$DISK \
			| parse-smart-attr \
			| sed "s/ERROR/$RES_ERROR/g" \
				> $OUTPUT_DIR/disk/$SERIAL.smart.error
		fi
	fi
	#RET=$?
	#echo $RET > $OUTPUT_DIR/disk/$SERIAL.fio.ret
	rm -f /run/$SERIAL.fio.pid
fi
END=`date +%s`

DURATION=$(( END - BEGIN ))

echo -n "${BLD}Process $MY_PROC${OFF} : " \
	> $OUTPUT_DIR/disk/$SERIAL.fio.result
if [[ -n $ERROR ]] ; then
	echo -n "Disk $DISK $SERIAL $ERROR : " \
		>> $OUTPUT_DIR/disk/$SERIAL.fio.result
	echo $RES_FAIL \
		>> $OUTPUT_DIR/disk/$SERIAL.fio.result	
	echo "$SERVER_SERIAL|$DISK|$SERIAL|FAIL|$ERROR" \
		> $OUTPUT_DIR/disk/$SERIAL.fio.csv
else
	echo -n "disk $DISK $SERIAL tested in ${DURATION}s : " \
		>> $OUTPUT_DIR/disk/$SERIAL.fio.result
	echo $RES_OK \
		>> $OUTPUT_DIR/disk/$SERIAL.fio.result
	echo "$SERVER_SERIAL|$DISK|$SERIAL|OK|${DURATION}s" \
		> $OUTPUT_DIR/disk/$SERIAL.fio.csv
fi

ln -sf $OUTPUT_DIR/disk/$SERIAL.fio.out \
	$OUTPUT_DIR/disk/$DISK.fio.out
ln -sf $OUTPUT_DIR/disk/$SERIAL.fio.result \
	$OUTPUT_DIR/disk/$DISK.fio.result
ln -sf $OUTPUT_DIR/disk/$SERIAL.fio.csv \
	$OUTPUT_DIR/disk/$DISK.fio.csv
if [[ -f $OUTPUT_DIR/disk/$SERIAL.smart.error ]] ; then
	ln -sf $OUTPUT_DIR/disk/$SERIAL.smart.error \
		$OUTPUT_DIR/disk/$DISK.smart.error
fi
	
cat $OUTPUT_DIR/disk/$SERIAL.fio.result
cat $OUTPUT_DIR/disk/$SERIAL.fio.result \
	>> $OUTPUT_DIR/fio.log