coop-computing-tools 4.0-1.1 / usr / bin / ec2_submit_workers

#!/bin/sh

show_help() 
{
        echo "Use: ec2_submit_workers [options] <servername> <port> <ec2-key-name> <ec2-key-file> <num-ec2-instances>"
        echo "where options are:"
        echo "  -M <name>           Name of the preferred master for worker."
        echo "  -N <name>           Same as -M (backwards compatibility)."
        echo "  -c <num>            Set the number of cores each worker should use (0=auto). (default=1)"
        echo "  -C <catalog>        Set catalog server to <catalog>. <catalog> format: HOSTNAME:PORT."
        echo "  -t <time>           Abort after this amount of idle time. (default=900s)."
        echo "  -d <subsystem>      Enable debugging on worker for this subsystem (try -d all to start)."
        echo "  -w <size>           Set TCP window size."
        echo "  -i <time>           Set initial value for backoff interval when worker fails to connect to a master. (default=1s)"
        echo "  -b <time>           Set maxmimum value for backoff interval when worker fails to connect to a master. (default=60s)"
        echo "  -z <size>           Set available disk space threshold (in MB). When exceeded worker will clean up and reconnect. (default=100MB)"
        echo "  -A <arch>           Set architecture string for the worker to report to master instead of the value in uname."
        echo "  -O <os>             Set operating system string for the worker to report to master instead of the value in uname."
        echo "  -s <path>           Set the location for creating the working directory of the worker."
        echo "  -n <number>         Set the number of workers to start on each EC2 instance. (default=1)"
        echo "  -I <image_id>       EC2 OS image ID. Default = ami-fa01f193 (Ubuntu 10.04 x86_64)."
        echo "  -Z <instance_size>  EC2 instance size. Default = m1.large."
        echo "  -p <parameters>     Set parameters for ec2-run-instances."
        echo "  -h                  Show this help message."
        exit 1
}

arguments=""
use_auto=0
image=ami-fa01f193
instance_size=m1.large
parameters=""
workers_per_instance=1

while getopts M:N:c:C:t:d:i:b:z:A:O:s:n:I:Z:p:h opt 
do
        case "$opt" in
                a)  arguments="$arguments -a"; use_auto=1;; #backwards compatibility
                M)  arguments="$arguments -M $OPTARG"; use_auto=1;;
                N)  arguments="$arguments -M $OPTARG"; use_auto=1;;
                c)  arguments="$arguments --cores $OPTARG"; cores=$OPTARG;;
                C)  arguments="$arguments -C $OPTARG";;
                t)  arguments="$arguments -t $OPTARG";;
                d)  arguments="$arguments -d $OPTARG";;
                w)  arguments="$arguments -w $OPTARG";;	 
                i)  arguments="$arguments -i $OPTARG";;	 
                b)  arguments="$arguments -b $OPTARG";;	 
                z)  arguments="$arguments -z $OPTARG";;	 
                A)  arguments="$arguments -A $OPTARG";;	 
                O)  arguments="$arguments -O $OPTARG";;	 
                s)  arguments="$arguments -s $OPTARG";;	 
                n)  workers_per_instance="$OPTARG";; 
                I)  image="$OPTARG";;
                Z)  instance_size="$OPTARG";;
                p)  parameters="$parameters $OPTARG";;
                h)  show_help;;
                \?) show_help;;
        esac
done

shift $(expr $OPTIND - 1)

if [ $use_auto = 0 ]; then
    if [ X$5 = X ]
    then
        show_help
    fi
    host=$1
    port=$2
    keyname=$3
    keyfile=$4
    count=$5
else
    if [ X$3 = X ]
    then
        show_help
    fi
    host=
    port=
    keyname=$1
    keyfile=$2
    count=$3
fi

worker=`which work_queue_worker 2>/dev/null`
if [ $? != 0 ]
then
        echo "$0: please add 'work_queue_worker' to your PATH."
        exit 1
fi

ec2run=`which ec2-run-instances 2>/dev/null`
if [ $? != 0 ]
then
        echo "$0: please add 'ec2-run-instances' to your PATH."
        exit 1
fi

cp $worker .

cat >worker.sh <<EOF
#!/bin/sh

for i in `eval echo {1..$workers_per_instance}`
 do
	./work_queue_worker $arguments $host $port &
done
EOF

chmod 755 worker.sh

#Start EC2 instances of the required count.
ec2run_out=$(ec2-run-instances $image --instance-type $instance_size -k $keyname $parameters -n $count)
reservation_id=$(echo $ec2run_out | awk '{print $2}')
echo "Created $count instance(s) under reservation $reservation_id."
sleep 60

#Wait until all of the requested instances come on-board..
running_instances=0
while [ $running_instances -lt $count ]
do
	sleep 20
	running_instances=$(ec2-describe-instances -F 'reservation-id'=$reservation_id | grep $image | grep running | wc -l)
done

declare -i inst_count
inst_count=1

while [ $inst_count -le $count ]
do
	instance=$(ec2-describe-instances -F 'reservation-id'=$reservation_id | grep running | awk 'NR==var{print $4 }' var="${inst_count}")
	echo "------------------------------------------------"
	echo "Starting worker #$inst_count on $instance"
	scp -o StrictHostKeyChecking=no -o ConnectTimeout=20 -o ConnectionAttempts=1 -i $keyfile $worker worker.sh ubuntu@$instance:.
	ssh -o StrictHostKeyChecking=no -o ConnectTimeout=20 -o ConnectionAttempts=1 -i $keyfile ubuntu@$instance './worker.sh 1>worker.log 2>&1 &'
	return_status=$?
	((inst_count=$inst_count+1))
done

exit $return_status