#!/bin/bash

# Use this instead of loadResultsD/loadResults when you want to make a single
# pass thru the running tasks db table to get results for jobs where there was
# a previous failed attempt.  Only considers tasks who's RunningTask table entry
# has status = TERMINATED (like loadResults) and retry > 1, meaning that we've
# already tried once.  This program will increment retry in the RunningTask record if
# there's an error.  Maximum retries is determined by the loadResults.retries property
# and code in sdk/tool/TaskMonitor.java.
#
# Looks for the working directory in the normal location as well as in the 
# FAILED and ARCHIVED subdirs.
#
# Can add a -Dlocal=path argument to the command line to look in a local filesystem instead
# of the directory where the task was run on a remote host.  NOT IMPLEMENTED.

HOST=`hostname`
COMMAND=org.ngbw.utils.LoadResults

SUBMITTER=submitter=${portal.url}
LOG=log4j.configuration=recoverResults_log4j.xml
PROPERTIES="-D$SUBMITTER -D$LOG -Drecover=1" 

RUN="python $SDK_VERSIONS/daemonize $PROPERTIES $COMMAND"

PS="ps aux -ww"

# Get processes IDs of interest
fn_ps()
{

	# We are looking for running instances of our COMMAND for same SUBMITTER.
	# If SUBMITTER is different, we don't worry about it.
	PS_OUTPUT=`$PS | grep $COMMAND | grep $SUBMITTER | grep recover | grep -v grep ` 
	PID=`echo "$PS_OUTPUT" |  awk '{print $2}'`
	len $PID ; PID_COUNT=$?

	# If the command is running, is someone else running it?  Set OTHER_USER to 0 or 1 depending
	# on whether another user is running the command we're trying to control.
	OTHER_USER=0
	USERS=`echo "$PS_OUTPUT" |  awk '{print $1}'`
	len $USERS ; USER_COUNT=$?
	if [ $USER_COUNT -ne 0 ]; then
		for u in $USERS
		do
			if [ "$u" != `whoami` ]; then
				OTHER_USER=1
			fi
		done
	fi
}


len()
{
	retval=0
	for i in $* 
	do
		retval=`expr $retval + 1`
	done
	return $retval 
}

killall()
{
	for pid in $*
	do
		echo killing $pid on $HOST
		kill $pid
	done
}



if [ "$SDK_VERSIONS" == "" ] ; then
	echo "SDK_VERSIONS environment variable must be set."
	exit 1
fi

case "$1" in
	'start')
		fn_ps
		if [ $PID_COUNT -eq 0 ]; then
			echo "Starting $COMMAND $PROPERTIES on $HOST"
			$RUN
			sleep 5
		elif [ $PID_COUNT -gt 0 ]; then 
			if [ $OTHER_USER -gt 0 ]; then
				echo "Another user is running $COMMAND $SUBMITTER on $HOST.  Pids are $PID".
				echo "$PS_OUTPUT"
			else
				echo "already running. pid is $PID  on $HOST"
			fi
		fi
		;;
	'stop')
		fn_ps
		killall $PID
		$0 status
		;;
	'restart')
		echo "Restarting $COMMAND $SUBMITTER " on $HOST
		$0 stop
		sleep 5
		$0 start
		;;
	'status')
		fn_ps
		if [ $PID_COUNT -gt 0 ] ; then
			if [ $OTHER_USER -gt 0 ]; then
				echo "Another user is running $COMMAND $SUBMITTER on $HOST.  Pids are $PID".
				echo "$PS_OUTPUT"
			else
				echo "$COMMAND $SUBMITTER is Running on $HOST.  pids: $PID"
			fi
		else
				echo "$COMMAND $SUBMITTER is NOT Running on $HOST"
		fi
		;;
		*)
			echo "Usage: $0 {start|stop|restart|status}"
			exit 0
esac