监控项目---磁盘Memory模块

监控项目---磁盘Memory模块

  磁盘IO模块涉及到的参数有:

    alarm_memory_up 20  可用内存的利用率%
    alarm_memory_times 3  报警次数

  Memory一共用到三个数据文件,分别为:mem_buffer mem_cache mem_free mem_total 利用率计算

    (buffer+cache+free)*100/total 取整跟alarm_memory_up比较..连续三次不足报警..

  脚本名:Module_Memory_Alarm.sh 内容如下:

    其它模块报警算法皆类似于此二模块,在此不作介绍了..如:LOAD CPU HTTP SQUID等及一些服务状况监控...

  #!/bin/sh
  if [ $# -ne 1 ];then
          /bin/echo "Usage: $0 IP"
          exit 1
  fi
 
  MONITOR_DIR=/performance_center/performance_monitor
  DEFAULT_CONF_DIR=$MONITOR_DIR/default_conf
  DATA_DIR=/cacti/idctraffic/hosttraffic
  CONF_DIR=$DEFAULT_CONF_DIR
 
  #check data validity
  IP=$1
  if [[ -f $DATA_DIR/$IP/mem_buffer && -f $DATA_DIR/$IP/mem_cache && -f $DATA_DIR/$IP/mem_free && -f $DATA_DIR/$IP/mem_total ]];then
          CURRENTLY_MEM_BUFFER=`/bin/cat $DATA_DIR/$IP/mem_buffer`
          CURRENTLY_MEM_CACHE=`/bin/cat $DATA_DIR/$IP/mem_cache`
          CURRENTLY_MEM_FREE=`/bin/cat $DATA_DIR/$IP/mem_free`
          CURRENTLY_MEM_TOTAL=`/bin/cat $DATA_DIR/$IP/mem_total`
  else
          /bin/echo $DATE "$IP memory idle is no data_file" >> $MONITOR_DIR/logs/Memory_error.log
          $MONITOR_DIR/scripts/Alarm.sh "$IP memory idle is no data_file" 2>>$MONITOR_DIR/logs/Memory_error.log 1>/dev/null
          exit 1
  fi
 
  CURRENTLY_MEMORY=`/bin/echo $CURRENTLY_MEM_BUFFER $CURRENTLY_MEM_CACHE $CURRENTLY_MEM_FREE $CURRENTLY_MEM_TOTAL| awk '{printf "%d", ($1+$2+$3)*100/$4}'`
  MD5_FILENAME_MEMORY=`/bin/echo "$IP memory idle is lack" -n | /usr/bin/md5sum | /bin/awk '{print $1}'`
 
  #set conf
  if [ -d $DATA_DIR/$IP/alarm_conf ];then
          CONF_DIR=$DATA_DIR/$IP/alarm_conf
          if [ -f $CONF_DIR/alarm_memory_times ];then
                  MEMORY_ALARM_TIMES=`/bin/cat $CONF_DIR/alarm_memory_times`
          else
                  MEMORY_ALARM_TIMES=`/bin/cat $DEFAULT_CONF_DIR/default.conf | /bin/grep "alarm_memory_times" | /bin/awk '{print $2}'`
          fi
          if [ -f $CONF_DIR/alarm_memory_up ];then
                  MEMORY_ALARM_VALUE_UP=`/bin/cat $CONF_DIR/alarm_memory_up`
          else
                  MEMORY_ALARM_VALUE_UP=`/bin/cat $DEFAULT_CONF_DIR/default.conf | /bin/grep "alarm_memory_up" | /bin/awk '{print $2}'`
          fi
  else
          MEMORY_ALARM_TIMES=`/bin/cat $DEFAULT_CONF_DIR/default.conf | /bin/grep "alarm_memory_times" | /bin/awk '{print $2}'`
          MEMORY_ALARM_VALUE_UP=`/bin/cat $DEFAULT_CONF_DIR/default.conf | /bin/grep "alarm_memory_up" | /bin/awk '{print $2}'`
  fi
 
  #alarm check
  if [ $CURRENTLY_MEMORY -lt $MEMORY_ALARM_VALUE_UP ];then
          if [ ! -f $MONITOR_DIR/alarm_info/$MD5_FILENAME_MEMORY ];then
                  /bin/echo 1 > $MONITOR_DIR/alarm_info/$MD5_FILENAME_MEMORY
                  /bin/echo "$IP memory idle is lack $MEMORY_ALARM_VALUE_UP:$CURRENTLY_MEMORY" >> $MONITOR_DIR/alarm_info/$MD5_FILENAME_MEMORY
          else
                  CUR_ALARM_TIMES_MEMORY=`/bin/cat $MONITOR_DIR/alarm_info/$MD5_FILENAME_MEMORY | /bin/sed -n '{1p}'`
                  CUR_ALARM_TIMES_MEMORY=`/usr/bin/expr $CUR_ALARM_TIMES_MEMORY + 1`
                  if [ $CUR_ALARM_TIMES_MEMORY -ge $MEMORY_ALARM_TIMES ];then
                          $MONITOR_DIR/scripts/Alarm.sh "$IP memory idle is lack $MEMORY_ALARM_VALUE_UP:$CURRENTLY_MEMORY" 2>$MONITOR_DIR/logs/Memory_error.log 1>/dev/null &
                          /bin/rm -rf $MONITOR_DIR/alarm_info/$MD5_FILENAME_MEMORY
                  else
                          /bin/echo $CUR_ALARM_TIMES_MEMORY > $MONITOR_DIR/alarm_info/$MD5_FILENAME_MEMORY
                          /bin/echo "$IP memory idle is lack $MEMORY_ALARM_VALUE_UP:$CURRENTLY_MEMORY" >> $MONITOR_DIR/alarm_info/$MD5_FILENAME_MEMORY
                  fi
          fi
  else
          if [ -f $MONITOR_DIR/alarm_info/$MD5_FILENAME_MEMORY ];then
                  /bin/rm -rf $MONITOR_DIR/alarm_info/$MD5_FILENAME_MEMORY
          fi
  fi
  

  exit 0