check_railo

#!/bin/bash

# Requests a testpage using a timeout, and restarts Railo (or Adobe ColdFusion server, 
# or something else) if the page doesn't respond. Also logs http status code if not "200 OK".
#
# USAGE: check_railo [testpage url]
#
# - Run as root
# - Special dependencies: curl
#
# Cron example (every three minutes during working hours):
# */3 7-18 * * * root  /usr/local/sbin/check_webserver http://localhost/testpage.cfm
#
# A suitable testpage is a cfm template that runs a simple database query on each datasource.
#

#### CONF SECTION ####
#
# Seconds to wait for testpage to answer
TIMEOUT="7" 
#
# Restart railo only after RESTART_INTERVAL minutes after last restart. Prevents restart loop.
RESTART_INTERVAL="60" 
#
# Logfile
LOGFILE="/var/log/check_railo.log"
#
# Command to run. Could also be "service coldfusion restart" or something else.
RESTART_COMMAND="/opt/coldfusion2016/cfusion/bin/coldfusion restart"
#
# Command for sendin email. Syntax as mailutils /usr/bin/mail, e.g. MAIL_COMMAND [subject] [to]
# Comment out for no mails
MAIL_COMMAND="/usr/local/sbin/telegrambot.py"
#
# Mail recipient
#ERROR_MAIL="$(cat /etc/scriptmail.txt)"
ERROR_MAIL="telegram"
#
#### END OF CONF SECTION ####



if [ $# -ne 1 ]
then
	echo "USAGE: check_railo [testpage url]"
	exit
fi

TESTPAGE=$1

killtimeout=$(($TIMEOUT+1)) # Kill command one second after timeout

timefile="/tmp/check_railo_timefile" # File for storing RESTART_INTERVAL
if [ ! -f "$timefile" ];
then
	# If missing (first run), create timefile and make it 12 hours old
	/usr/bin/touch $timefile -d '-12 hour'
fi
timefile_age=$(( (`date +%s` - `stat -L --format %Y $timefile`)/60 )) # Get file age in minutes

# Run the test and save the result.
RESULT=$(/usr/bin/timeout -k $killtimeout $TIMEOUT /usr/bin/curl -s -o /dev/null -I -w "%{http_code}" ${TESTPAGE})

if [ "$RESULT" == "" ];
then
	ERRORSTATUS="TIMEOUT"

	if [ "$timefile_age" -gt "$RESTART_INTERVAL" ];
	then
		ERRORSTATUS="TIMEOUT: INIT RESTART"
		/usr/bin/touch $timefile
		# Restart server
		$RESTART_COMMAND
	else
		# Don't restart until timefile is old enough
		ERRORSTATUS="TIMEOUT: LAST RESTART TOO RECENT ($timefile_age min)"
	fi


elif [ "$RESULT" != 200 ];
then
	# Responds in time but returns http error: save error for logging.
	ERRORSTATUS=$RESULT
fi

if [ "$ERRORSTATUS" ];
then
	if [ "$MAIL_COMMAND" ];
	then
		# Send mail
		$MAIL_COMMAND "(check_railo) ERROR: $ERRORSTATUS" "$ERROR_MAIL"
		ERRORSTATUS="$ERRORSTATUS, mail sent to $ERROR_MAIL"
	fi

	# Log errorstatus
	echo "$(/bin/date "+%Y-%m-%d %H:%M:%S") ERROR: $ERRORSTATUS" >> $LOGFILE
fi