resin的自动拉起脚本，检测僵死进程

#! /bin/sh
#
# See contrib/init.resin for /etc/rc.d/init.d startup script
#
# resin.sh can be called like apachectl
#
# resin.sh         -- execs resin in the foreground
# resin.sh start   -- starts resin in the background
# resin.sh stop    -- stops resin
# resin.sh restart -- restarts resin
#
# resin.sh will return a status code if the wrapper detects an error, but
# some errors, like bind exceptions or Java errors, are not detected.
#
# To install, you'll need to configure JAVA_HOME and RESIN_HOME and
# copy contrib/init.resin to /etc/rc.d/init.d/resin.  Then
# use "unix# /sbin/chkconfig resin on"

PCK_TAG=pcheck
export=PCK_TAG

#JAVA_HOME
JAVA_HOME=/usr/local/jdk1.6.0_29
export JAVA_HOME
#本脚本的绝对路径
EXEC_START="/usr/local/sns/snsconnector.sh"

#日志输入路径
WATCH_LOG="/usr/local/sns/watchdog.log"
export WATCH_LOG

#执行脚本
RESIN="bin/httpd.sh"
#执行多个resin的多路径
EXEC="/usr/local/sns/snsapi/resin3 /usr/local/sns/snsapi2/resin3"

array=($EXEC)
length=${#array[@]}
echo $length
for ((i=0; i<$length; i++))
do
    echo ${array[$i]}
done

#####################################################
# Find a PID for the pid file
#####################################################

for ((i=0; i<$length; i++))
do
	RESIN_HOME=${array[$i]}
	SNS_PIDS="$SNS_PIDS $RESIN_HOME/snsconnector.pid "
done

pids=($SNS_PIDS)
size=${#pids[@]}

echo $size
for ((i=0; i<$size; i++))
do
    echo ......${pids[$i]}
done


#
SNS_TAG=snstag


if test -n "${JAVA_HOME}"; then
  if test -z "${JAVA_EXE}"; then
    JAVA_EXE=$JAVA_HOME/bin/java
  fi
fi  

#
# trace script and simlinks to find the wrapper
#
if test -z "${RESIN_HOME}"; then
  script=`/bin/ls -l $0 | awk '{ print $NF; }'`

  while test -h "$script"
  do
    script=`/bin/ls -l $script | awk '{ print $NF; }'`
  done

  bin=`dirname $script`
  RESIN_HOME="$bin/.."
fi  

running()
{
    [ -f $1 ] || return 1
    PID=$(cat $1)
    echo ==========$PID==========
    ps -p $PID >/dev/null 2>/dev/null || return 1
    return 0
}

zombiling()
{
    PID=$(cat $1)
    echo "zombiling $PID" >>$WATCH_LOG
    IMZOMBIE=`ps -el |grep $PID|awk '{print $2}'` >/dev/null 2>/dev/null 
	imarray=($IMZOMBIE)
	echo "SNSConnector process current run status is: ${imarray[0]}" >>$WATCH_LOG
	echo "SNSConnector process current run status is: $IMZOMBIE"
	echo "`date '+%Y-%m-%d %T'` [INFO] SNSConnector process current run status is: $IMZOMBIE" >>$WATCH_LOG
	if [ ! -z `echo ${imarray[0]}|grep '^[TZ]'` ]
	then
		echo "SNSConnector process is zombied. now restart the service."
		echo "`date '+%Y-%m-%d %T'` [WARN] SNSConnector process is zombied. now restart the service." >>$WATCH_LOG
		return 0
	fi
    return 1
}

usage()
{
    echo "Usage: $0 {start|stop|status} [ CONFIGS ... ] "
    exit 1
}

[ $# -gt 0 ] || usage


ACTION=$1
MAINTE_CHECK=$2
ZOMBIE_CHECK=$3

#####################################################
# Get current script belongs to User and Group.
#####################################################
[ -z ${SCRIPT_NAME} ] && SCRIPT_NAME=httpd.sh
if [ -f ${RESIN_HOME}/bin/${SCRIPT_NAME} ]
then
	CURRENT_GROUP="$(echo `ls -l ${RESIN_HOME}/bin/${SCRIPT_NAME}` | awk '{print $4}')";
	CURRENT_USER="$(echo `ls -l ${RESIN_HOME}/bin/${SCRIPT_NAME}` | awk '{print $3}')";
else
	echo "`date '+%Y-%m-%d %T'` ERROR: Current start service script is:${RESIN_HOME}/bin/${SCRIPT_NAME}" >>$WATCH_LOG
	echo "`date '+%Y-%m-%d %T'` ERROR: The SNS start file is not exist."
fi
setCrontab()
{
    echo "setCrontab...";
    crontab_tmp=$(mktemp -p "${RESIN_HOME}/bin/")



		if [ x$CURRENT_USER != x$(whoami) ] 
		then
			echo "* * * * * su - $CURRENT_USER -c "${EXEC_START} start $SNS_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
			echo "* * * * * sleep 10; su - $CURRENT_USER -c "${EXEC_START} start $SNS_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
			echo "* * * * * sleep 20; su - $CURRENT_USER -c "${EXEC_START} start $SNS_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
			echo "* * * * * sleep 30; su - $CURRENT_USER -c "${EXEC_START} start $SNS_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
			echo "* * * * * sleep 40; su - $CURRENT_USER -c "${EXEC_START} start $SNS_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
			echo "* * * * * sleep 50; su - $CURRENT_USER -c "${EXEC_START} start $SNS_TAG $PCK_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
		elif [ x$CURRENT_USER == x$(whoami) ]
		then
			echo "* * * * * "${EXEC_START} start $SNS_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
			echo "* * * * * sleep 10; "${EXEC_START} start $SNS_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
			echo "* * * * * sleep 20; "${EXEC_START} start $SNS_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
			echo "* * * * * sleep 30; "${EXEC_START} start $SNS_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
			echo "* * * * * sleep 40; "${EXEC_START} start $SNS_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
			echo "* * * * * sleep 50; "${EXEC_START} start $SNS_TAG $PCK_TAG" > /dev/null 2>&1" >> "${crontab_tmp}"
		fi
	
	tail "${crontab_tmp}"
	crontab "${crontab_tmp}"
	rm -f ${crontab_tmp}
	return 0
}


case "$ACTION" in
  start)
	echo "`date '+%Y-%m-%d %T'` [INFO] snsconnector service start begin.$MAINTE_CHECK $ZOMBIE_CHECK " >>$WATCH_LOG
	for ((i=0; i<$length; i++))
	do
		RESIN_HOME=${array[$i]}
		RUN_CMD=$RESIN_HOME/$RESIN
		SNS_PID=${pids[$i]}
		echo -n "$RESIN_HOME Starting snsconnector: "
		if [ x$MAINTE_CHECK == x$SNS_TAG ] && [ -f "$RESIN_HOME/$SNS_TAG" ]
		then
			echo "Maintenance period,can't start sns service."
			exit 1
		fi
		

		echo "`date '+%Y-%m-%d %T'` [INFO] snsconnector service start begin.[x$ZOMBIE_CHECK] [x$PCK_TAG] " >>$WATCH_LOG
		if [ x$ZOMBIE_CHECK == x$PCK_TAG ] && zombiling $SNS_PID
		then
			PID=`cat $SNS_PID 2>/dev/null`
			echo "`date '+%Y-%m-%d %T'` snsconnector(pid $PID $SNS_PID) is zombie." >>$WATCH_LOG
			kill -9 $PID
			echo "`date '+%Y-%m-%d %T'` snsconnector(pid $SNS_PID) is stopped." >>$WATCH_LOG
			continue
		else
			if [ x$ZOMBIE_CHECK == x$PCK_TAG ] 
			then
				echo "snsconnector(pid `cat $SNS_PID`) is not zombie." >>$WATCH_LOG
			fi
		fi
		echo "`date '+%Y-%m-%d %T'` [INFO] >>>>>>>>>>[x$ZOMBIE_CHECK] [x$PCK_TAG] " >>$WATCH_LOG

		if [ -f $SNS_PID ]
		then
			if running $SNS_PID
			then
				echo "Already Running."
				echo "`date '+%Y-%m-%d %T'` snsconnector running pid is:"`cat $RESIN_HOME/snsconnector.pid` >>$WATCH_LOG
				#exit 1
				continue
			else
				# dead pid file - remove
				rm -f $SNS_PID
			fi
		fi
	    
		if [ x$CURRENT_USER != x ] && [ x$CURRENT_USER != x$(whoami) ] 
		then
			touch $SNS_PID
			chown $CURRENT_USER:$CURRENT_GROUP $SNS_PID
			su - $CURRENT_USER -c "
			$RUN_CMD > /dev/null 2>&1 &
			PID=\$!
			disown \$PID
			echo \$PID > $SNS_PID"
			echo pid=$PID
			echo sns_pid=$SNS_PID
		elif [ x$CURRENT_USER == x$(whoami) ]
		then
			echo ===$RUN_CMD===
			touch $SNS_PID
			eval "
			$RUN_CMD > /dev/null 2>&1 &
			PID=\$!
			disown \$PID
			echo \$PID > $SNS_PID"
			echo pid=$PID
                        echo sns_pid=$SNS_PID
		fi 
		sleep 5
		if running $SNS_PID
		then
			rm -f $RESIN_HOME/$SNS_TAG
			echo "OK"
			echo "pid is `cat $SNS_PID 2>/dev/null`"
			echo "STARTED AT `date`"
			echo "`date '+%Y-%m-%d %T'` [INFO] snsconnector service start OK." >>$WATCH_LOG
		else
			echo "FAILED."
			 echo "`date '+%Y-%m-%d %T'` [ERROR] snsonnector service start FAILED." >>$WATCH_LOG
		fi
	done
	setCrontab;
	exit 1
        ;;

  stop)

	echo "`date '+%Y-%m-%d %T'` [INFO] snsconnector service stop begin." >>$WATCH_LOG
	for ((i=0; i<$length; i++))
	do
		RESIN_HOME=${array[$i]}
		SNS_PID=${pids[$i]}
		echo -n "$RESIN_HOME Stopping snsconnector... "
		if running $SNS_PID
		then
			PID=`cat $RESIN_HOME/$SNS_PID 2>/dev/null`
			TIMEOUT=60
			while running $SNS_PID && [ $TIMEOUT -gt 0 ]
			do
				kill $PID 2>/dev/null
				echo -n "."
				sleep 1
				let TIMEOUT=$TIMEOUT-1
			done
		
			[ $TIMEOUT -gt 0 ] || kill -9 $PID 2>/dev/null
			rm -f $SNS_PID
			echo OK
			echo "`date '+%Y-%m-%d %T'` [INFO] $RESIN_HOME snsconnector service stop OK." >>$WATCH_LOG
		else
			echo "$SNS_PID snsconnector service already stopped,stop failed."
			echo "`date '+%Y-%m-%d %T'` [ERROR] $RESIN_HOME snsconnector service already stopped,stop FAILED." >>$WATCH_LOG
		fi
	        touch $RESIN_HOME/$SNS_TAG
	done
	exit 1
        ;;

  status)
	echo "`date '+%Y-%m-%d %T'` [INFO] snsconnector service status." >>$WATCH_LOG
	for ((i=0; i<$length; i++))
	do
	    RESIN_HOME=${array[$i]}
	    SNS_PID=${pids[$i]}
		    if running $SNS_PID
		    then
			PID=`cat $SNS_PID 2>/dev/null`
			echo "snsconnector(pid $PID) is running."
			echo "`date '+%Y-%m-%d %T'` [INFO] $RESIN_HOME snsconnector(pid $PID) is running." >>$WATCH_LOG
		    else
			rm -f $SNS_PID
			echo "snsconnector has stoped."
			echo "`date '+%Y-%m-%d %T'` [INFO] $RESIN_HOME snsconnector(pid $PID) has stoped." >>$WATCH_LOG
		    fi
	done
    exit 1
    ;;

*)
        usage
    ;;
esac

exit 0