#!/bin/bash -u

CONFIG_FILE=gf_perf_config
source $CONFIG_FILE

SETTLE_TIME=10
RUNFILE=.runfile
STAT_COLLECTOR=stat_collect

# Generate current run, update runfile
if [ ! -f $RUNFILE ]
then
        run=1
else
        run=`cat $RUNFILE`
fi
echo $((run+1)) > $RUNFILE

# Drop vm caches on all the bricks before starting the runs
for brick in $BRICK_IP_ADDRS
do
        ssh -l root $brick "echo 3 > /proc/sys/vm/drop_caches"
done

# Create the gluster volume
./create_gluster_vol

mount | grep $MOUNT_POINT > /dev/null 2>&1
if [ $? -eq 0 ]
then
	umount $MOUNT_POINT
fi

if [ ! -d $MOUNT_POINT ]
then
	mkdir -p $MOUNT_POINT
fi

# Make sure that the fuse kernel module is loaded
/sbin/lsmod | grep -w fuse > /dev/null 2>&1
if [ $? -ne 0 ]
then
	/sbin/modprobe fuse > /dev/null 2>&1
fi

ps -eaf | egrep -w 'glusterfs|glusterfsd|glusterd' > /dev/null 2>&1
if [ $? -eq 0 ]
then
	killall glusterfsd glusterd glusterfs  > /dev/null 2>&1
fi

# Mount the client
# Sleep for a while. Sometimes, NFS mounts fail if attempted soon after creating the volume
sleep $SETTLE_TIME

if [ $ENABLE_ACL == "yes" ]
then
        acl_opts="-o acl"
else
        acl_opts=""
fi

echo "Mounting volume..."
if [ $ENABLE_MEM_ACCT == "yes" ]
then
	echo "Memory accounting status on client -"
	echo "x/x &gf_mem_acct_enable" > commands.$$
	echo "quit" >> commands.$$
	GLUSTERFS_DISABLE_MEM_ACCT=0 mount -t $MOUNT_TYPE $acl_opts $MGMT_NODE:$VOLNAME $MOUNT_POINT
	mount_status=$?
	gdb -q --command=commands.$$ -p `pidof glusterfs` | grep gf_mem_acct_enable | awk '{print $(NF-1) $NF}'
	rm commands.$$ > /dev/null 2>&1
else
	mount -t $MOUNT_TYPE $acl_opts $MGMT_NODE:$VOLNAME $MOUNT_POINT
	mount_status=$?
fi

if [ $mount_status -ne 0 ]
then
        echo "mount -t $MOUNT_TYPE $acl_opts $MGMT_NODE:$VOLNAME $MOUNT_POINT failed..."
        echo "Exiting..."
        exit 1
fi

# Copy statistics collection scripts to the server

echo ""
echo "Copying stat collection script to bricks..."
for brick in $BRICK_IP_ADDRS
do 
	ssh -l root $brick "mkdir -p $SERVER_SCRIPTS_DIR"
	scp -p $STAT_COLLECTOR root@$brick:$SERVER_SCRIPTS_DIR > /dev/null 2>&1
done

# Run statistics collection scripts on the server

echo ""
echo "Starting server stat collection..."
for brick in $BRICK_IP_ADDRS
do 
	ssh -l root $brick "mkdir -p $SERVER_LOG_DIR"
	ssh -l root $brick "$SERVER_SCRIPTS_DIR/$STAT_COLLECTOR $SERVER_LOG_DIR" &
done

# Run statistics collection on client

mkdir -p $LOCAL_LOG_REPO
./$STAT_COLLECTOR $LOCAL_LOG_REPO/run$run/client &

# Start perf test

echo ""
echo "Starting run $run..."

sleep $SETTLE_TIME
./parallel_create 50 10000
sleep $SETTLE_TIME

# Stop statistics collection scripts on the client

killall mpstat vmstat iostat $STAT_COLLECTOR sar > /dev/null 2>&1

# Stop statistics collection scripts on the server

echo ""
echo "Stopping server stat collection..."
for brick in $BRICK_IP_ADDRS
do 
	ssh -l root $brick killall mpstat vmstat iostat $STAT_COLLECTOR sar > /dev/null 2>&1
done

# Since the ssh processes were backgrounded, they will be hanging around.
# Kill them
kill `jobs -l | awk '{print $2}'` > /dev/null 2>&1

# Copy statistics from the server

echo ""
echo "Copying server logfiles for run $run..."
cur_log_dump_dir=$LOCAL_LOG_REPO/run$run
mkdir -p $cur_log_dump_dir
count=1
for brick in $BRICK_IP_ADDRS
do
        for statf in mpstat vmstat iostat sysinfo sar_netstat
        do
                scp root@$brick:$SERVER_LOG_DIR/*$statf* $cur_log_dump_dir/brick$count-$brick-$statf-log > /dev/null 2>&1
        done
        count=$((count + 1))
done

# Cleanup statistics collected for this run on the server

echo ""
echo "Cleaning server logfiles..."
for brick in $BRICK_IP_ADDRS
do 
	ssh -l root $brick "cd $SERVER_LOG_DIR; rm mpstat_log vmstat_log iostat_log sysinfo sar_netstat_log"
done