#!/bin/bash
#
# Init file for Parasol server daemon
#
# description: Parasol server daemon
#
# processname: parasol

# no shell limits to interfere
ulimit -c unlimited

# directory where all parasol binaries are available
export binDir="/data/bin"
# directory to run the command in and were log files will accumulate
export workDir="/data/parasol"

# the internal cloud address of this machine:
export parasolHub=`ip addr show | egrep "inet.*eth0|inet.*em1" | head -1 | awk '{print $2}' | sed -e 's#/.*##;'`

# date stamp for log identification
export dateStamp=`date "+%Y-%m-%dT%H:%M"`
# assume exit code is zero
export exitCode=0
export logFile="${workDir}/logs/${parasolHub}.${dateStamp}.log"
export nodeList=${parasolHub}.ms
# if the para node process do not report back on completed jobs, this
# subNet may be too restrictive
export subNet=`echo $parasolHub | awk -F'.' '{printf "%d.%d", $1,$2}'`

# using ssh to communicate with the node machines
export RSH="-rsh=/usr/bin/ssh"

# from here through the rest of the script, assume working in this directory
cd "${workDir}"
mkdir -p logs

# the < /dev/null to the ssh command in the while loop prevents this loop
#  from running only once.  Something about consuming stdin makes the
#  loop stop after only one pass

# verify all the ssh keys, make their entries in known_hosts so there is
# no pausing during para start
initSshKeys()
{
	cat nodeInfo/*.ms > ${nodeList}
        cut -f1 ${nodeList} | while read ip
do
   ssh -x -o 'StrictHostKeyChecking = no' -o 'BatchMode = yes' $ip 'date' < /dev/null > /dev/null
   printf "# initialized ssh to node: $ip\n" 1>&2
done
}

start()
{
	echo -n $"Starting parasol:"
        rm -fr /dev/shm/parasol
        mkdir -p /dev/shm/parasol
	cat nodeInfo/*.ms > ${nodeList}
	${binDir}/paraNodeStart ${RSH} $nodeList -hub=${parasolHub} -log=${logFile}.node -umask=002 -userPath=bin -sysPath=. -exe=${binDir}/paraNode
	${binDir}/paraHub -log=${logFile}.hub $nodeList subnet=$subNet &
	echo Done.
}

restartHub()
{
	echo -n "Restarting just paraHub:"
        echo Stopping Hub...
	${binDir}/paraHubStop now
        echo Restarting Hub...
	rm -fr /dev/shm/parasol
	mkdir -p /dev/shm/parasol
	cat nodeInfo/*.ms > ${nodeList}
	${binDir}/paraHub -log=${logFile}.hub $nodeList subnet=$subNet &
        echo Done.
}

stop()
{
	echo -n $"Stopping parasol:"
	${binDir}/paraNodeStop $nodeList
	${binDir}/paraHubStop now
        rm -fr /dev/shm/parasol
}

resurrect()
{
	echo -n $"Resurrect dead parasol nodes:"
	rm -f dead.list
        ${binDir}/parasol list machines | grep dead | awk '{print $1}' | sort -u | while read ip
do
 cat nodeInfo/$ip.ms
 ${binDir}/parasol remove machine "${ip}" "dead machine" > /dev/null 2>&1
done > dead.list
	printf "${binDir}/paraNodeStart ${RSH} dead.list -hub=${parasolHub} -log=${logFile}.node -umask=002 -userPath=bin -sysPath=. -exe=${binDir}/paraNode\n" 1>&2
	${binDir}/paraNodeStart ${RSH} dead.list -hub=${parasolHub} -log=${logFile}.node -umask=002 -userPath=bin -sysPath=. -exe=${binDir}/paraNode
        awk '{print $1}' dead.list | while read ms
do
  sudo ${binDir}/parasol add machine `cat "nodeInfo/${ms}.ms"`
done
}

case "$1" in
	start)
		start
		exitCode=$?
		;;
	stop)
		stop
		exitCode=$?
		;;
	restart)
		stop
		start
		exitCode=$?
		;;
	restartHub)
		restartHub
		exitCode=$?
		;;
	resurrect)
		resurrect
		exitCode=$?
		;;
	initialize)
		initSshKeys
		;;
	status)
		${binDir}/parasol status
		exitCode=$?
		;;
	*)
		echo $"Usage: $0 {initialize|start|stop|restart|restartHub|resurrect|status}"
		exitCode=1
esac
exit $exitCode