应用服务监控脚本用shell写的

#!/bin/bash PATH=/usr/local/bin:/usr/local/sbin:/usr/bin:/usr/sbin:/bin:/sbin ROOT=/data/dv_mon/dv_appchk mail_file() { # usage: mail_file "SUBJECT" MAIL_FILE SUBJECT=$1 MAIL_FILE=$2 [[ ! -f $MAIL_CONF ]] && return . $MAIL_CONF env MAILRC=/dev/null from="$SENDER" smtp="$SMTP_SVR" \ smtp-auth-user="$SMTP_USR" smtp-auth-password="$SMTP_PSW" smtp-auth=login \ nail -n -s "$SUBJECT" "$TO_WHO" < $MAIL_FILE } pre_chk() { echo "[$(date +%c)] pre_chk(): Begin." >> $LOG0 LOG1=$LOG0.1 LOG2=$LOG0.2 LOG3=$LOG0.3 MAIL_TMP=$RUN/mail.tmp cat /dev/null > $MAIL_TMP touch $HIS_LOG # CHK_RESULT: {_OK|_RESTART_OK|_START_FAIL|_STOP_FAIL} CHK_RESULT="" NEED_MAIL=no # get hostname HOST_NAME=$(hostname | awk -F'.' '{print $1}') # get ip address for iface in eth0 eth0:1 eth1 do ETH_IF=$(ifconfig $iface | awk -F: '/inet addr/{split($2,a," ") ; print a[1]}') [[ ! -z $ETH_IF ]] && break done # exit if can not get hostname or ip address if [[ -z $HOST_NAME || -z $ETH_IF ]] then echo "[$(date +%c)] pre_backup(): Can not get hostname or IP address!" >> $LOG0 exit fi # check lock file if [[ ! -f $LOCK_FILE ]] then touch $LOCK_FILE else LOCK_FILE_TIME=$(ls -l --time-style=+%s $LOCK_FILE | awk '{print $6}') NOW_TIME=$(date +%s) if (( NOW_TIME - LOCK_FILE_TIME < LOCK_FILE_TIMEOUT )) then echo "[$(date +%c)] pre_chk(): Check is still running. Exit now." >> $LOG0 exit else echo "[$(date +%c)] pre_chk(): Last check out of timeout $LOCK_FILE_TIMEOUT. Check now." >> $LOG0 rm -f $LOCK_FILE fi fi } chk_port() { echo "[$(date +%c)] chk_port(): Begin." >> $LOG0 netstat -tunlp > $RUN/netstat.tmp 2>&1 grep "\:$APP_PORT\>" $RUN/netstat.tmp > /dev/null 2>&1 ret=$? if (( ret == 0 )) then PORT_STATUS="up" else echo "quit" | telnet localhost $APP_PORT | col -b | grep "Connected" > /dev/null 2>&1 ret=$? if (( ret == 0 )) then PORT_STATUS="up" else sleep 3 echo "quit" | telnet localhost $APP_PORT | col -b | grep "Connected" > /dev/null 2>&1 ret=$? if (( ret == 0 )) then PORT_STATUS="up" else PORT_STATUS="down" echo "[$(date +%c)] chk_port(): PORT_STATUS = $PORT_STATUS" >> $MAIL_TMP cat $RUN/netstat.tmp >> $MAIL_TMP fi fi fi echo "[$(date +%c)] chk_port(): Return PORT_STATUS = $PORT_STATUS." >> $LOG0 rm -f $RUN/netstat.tmp } chk_proc() { echo "[$(date +%c)] chk_proc(): Begin." >> $LOG0 CHK_URL="http://localhost:${APP_PORT}${CHK_URI}" WGET_OPT="-v -d -t 2 -O /dev/null -S -T 8 -w 5 --proxy=off" wget $WGET_OPT "$CHK_URL" 2>&1 | col -b > $RUN/wget.tmp grep '200 OK' $RUN/wget.tmp > /dev/null 2>&1 ret=$? if (( ret == 0 )) then PROC_STATUS="up" else sleep 3 wget $WGET_OPT "$CHK_URL" 2>&1 | col -b > $RUN/wget.tmp.1 grep '200 OK' $RUN/wget.tmp.1 > /dev/null 2>&1 ret=$? if (( ret == 0 )) then PROC_STATUS="up" else PROC_STATUS="down" echo "[$(date +%c)] chk_proc(): PROC_STATUS = $PROC_STATUS" >> $MAIL_TMP cat $RUN/wget.tmp $RUN/wget.tmp.1 >> $MAIL_TMP fi fi echo "[$(date +%c)] chk_proc(): Return PROC_STATUS = $PROC_STATUS." >> $LOG0 rm -f $RUN/wget.tmp $RUN/wget.tmp.1 } start_app() { echo "[$(date +%c)] start_app(): Begin." >> $LOG0 $APP_CTRL start > /dev/null 2>&1 sleep 3 } stop_app() { echo "[$(date +%c)] stop_app(): Begin." >> $LOG0 $APP_CTRL stop > /dev/null 2>&1 sleep 3 } post_chk() { if [[ $NEED_MAIL == yes ]] then mail_file "APPCHK_${HOST_NAME}[$ETH_IF]:$CHK_RESULT" $MAIL_TMP fi # delete mail temp file rm -f $MAIL_TMP # delete lock file rm -f $LOCK_FILE } main_chk() { echo "[$(date +%c)] main_chk(): Begin." >> $LOG0 case "${CHK_PORT}_${CHK_PROC}" in enable_enable) echo "[$(date +%c)] main_chk(): CHK_PORT = $CHK_PORT, CHK_PROC = $CHK_PROC." >> $LOG0 chk_port chk_proc if [[ $PORT_STATUS == up && $PROC_STATUS == up ]] then # Status OK echo "[$(date +%c)] main_chk(): PORT_STATUS = $PORT_STATUS, PROC_STATUS = $PROC_STATUS." >> $LOG0 echo "[$(date +%c)] main_chk(): $APP_NAME status OK." >> $LOG0 else # Port down or process down stop_app chk_port chk_proc if [[ $PORT_STATUS == down && $PROC_STATUS == down ]] then # Stop application success start_app chk_port chk_proc if [[ $PORT_STATUS == up && $PROC_STATUS == up ]] then # Start application success NEED_MAIL=yes CHK_RESULT="${APP_NAME}_port_${APP_PORT}_RESTART_OK:${CHK_RESULT}" echo "[$(date +%c)] APP_NAME:$APP_NAME APP_PORT:$APP_PORT RESTART_OK" | tee -a $MAIL_TMP >> $HIS_LOG echo "[$(date +%c)] main_chk(): PORT_STATUS = $PORT_STATUS, PROC_STATUS = $PROC_STATUS." >> $LOG0 echo "[$(date +%c)] main_chk(): Restart $APP_NAME success." >> $LOG0 else # Fail to start application NEED_MAIL=yes CHK_RESULT="${APP_NAME}_port_${APP_PORT}_START_FAIL:${CHK_RESULT}" echo "[$(date +%c)] APP_NAME:$APP_NAME APP_PORT:$APP_PORT START_FAIL" | tee -a $MAIL_TMP >> $HIS_LOG echo "[$(date +%c)] main_chk(): PORT_STATUS = $PORT_STATUS, PROC_STATUS = $PROC_STATUS." >> $LOG0 echo "[$(date +%c)] main_chk(): Can not start $APP_NAME." >> $LOG0 fi else # Fail to stop application NEED_MAIL=yes CHK_RESULT="${APP_NAME}_port_${APP_PORT}_STOP_FAIL:${CHK_RESULT}" echo "[$(date +%c)] APP_NAME:$APP_NAME APP_PORT:$APP_PORT STOP_FAIL" | tee -a $MAIL_TMP >> $HIS_LOG echo "[$(date +%c)] main_chk(): PORT_STATUS = $PORT_STATUS, PROC_STATUS = $PROC_STATUS." >> $LOG0 echo "[$(date +%c)] main_chk(): Can not stop $APP_NAME." >> $LOG0 fi fi ;; disable_disable) echo "[$(date +%c)] main_chk(): CHK_PORT = $CHK_PORT, CHK_PROC = $CHK_PROC." >> $LOG0 echo "[$(date +%c)] main_chk(): Check disabled $APP_NAME." >> $LOG0 ;; enable_disable|disable_enable) echo "[$(date +%c)] main_chk(): CHK_PORT = $CHK_PORT, CHK_PROC = $CHK_PROC." >> $LOG0 [[ $CHK_PORT == enable && $CHK_PROC == disable ]] && CHK_LOGIC=chk_port && CHK_STATUS=PORT_STATUS [[ $CHK_PORT == disable && $CHK_PROC == enable ]] && CHK_LOGIC=chk_proc && CHK_STATUS=PROC_STATUS $CHK_LOGIC REAL_STATUS=$(eval echo '$'$CHK_STATUS) if [[ $REAL_STATUS == up ]] then # Status OK echo "[$(date +%c)] main_chk(): $CHK_STATUS = $REAL_STATUS." >> $LOG0 echo "[$(date +%c)] main_chk(): $APP_NAME status OK." >> $LOG0 else # Port down or process down stop_app $CHK_LOGIC REAL_STATUS=$(eval echo '$'$CHK_STATUS) if [[ $REAL_STATUS == up ]] then # Fail to stop application NEED_MAIL=yes CHK_RESULT="${APP_NAME}_port_${APP_PORT}_STOP_FAIL:${CHK_RESULT}" echo "[$(date +%c)] APP_NAME:$APP_NAME APP_PORT:$APP_PORT STOP_FAIL" | tee -a $MAIL_TMP >> $HIS_LOG echo "[$(date +%c)] main_chk(): $CHK_STATUS = $REAL_STATUS." >> $LOG0 echo "[$(date +%c)] main_chk(): Can not stop $APP_NAME." >> $LOG0 else # Stop application success start_app $CHK_LOGIC REAL_STATUS=$(eval echo '$'$CHK_STATUS) if [[ $REAL_STATUS == up ]] then # Start application success NEED_MAIL=yes CHK_RESULT="${APP_NAME}_port_${APP_PORT}_RESTART_OK:${CHK_RESULT}:" echo "[$(date +%c)] APP_NAME:$APP_NAME APP_PORT:$APP_PORT RESTART_OK" | tee -a $MAIL_TMP >> $HIS_LOG echo "[$(date +%c)] main_chk(): $CHK_STATUS = $REAL_STATUS." >> $LOG0 echo "[$(date +%c)] main_chk(): Restart $APP_NAME success." >> $LOG0 else # Fail to start application NEED_MAIL=yes CHK_RESULT="${APP_NAME}_port_${APP_PORT}_START_FAIL:${CHK_RESULT}" echo "[$(date +%c)] APP_NAME:$APP_NAME APP_PORT:$APP_PORT START_FAIL" | tee -a $MAIL_TMP >> $HIS_LOG echo "[$(date +%c)] main_chk(): $CHK_STATUS = $REAL_STATUS." >> $LOG0 echo "[$(date +%c)] main_chk(): Can not start $APP_NAME." >> $LOG0 fi fi fi ;; *) echo "[$(date +%c)] main_chk(): CHK_PORT = $CHK_PORT, CHK_PROC = $CHK_PROC." >> $LOG0 echo "[$(date +%c)] main_chk(): Not