1
#!/bin/bash
2

3
EMAIL_CONF=""
4
LOGFILE=""
5
CONF=""
6
ipAddress=$(curl -s ifconfig.me)
7
function sendEmail(){
8
    local userEmails=$(cat $EMAIL_CONF | jq -r 'to_entries[] | .value.email')
9
    local issue=$1
10
    local title=$2
11
    for email in $userEmails; do
12
        echo -e "时间: $(date +"%Y-%m-%d %H:%M:%S")\n主机IP: ${ipAddress}\n问题: $issue" | mail -s "$title" "$email"
13
    done
14
}
15
function writeLog(){
16
    echo -e "$(date +"%Y-%m-%d %H:%M:%S") - $1" >> $LOGFILE
17
}
18
function cpuUsage(){
19
    local cpuIdle=$(top -bn1 | grep Cpu | sed 's/,/ /g'|awk '{print $8}' | cut -d. -f1)
20
    local cpuUsed=$[100-$cpuIdle]
21
    local cpuLimit=$(jq '.cpu.cpu_limit' $CONF)
22
    #echo "cpuLimit: $cpuLimit"
23
    if [ $cpuUsed -gt $cpuLimit ]; then
24
        sendEmail "CPU使用率过高 - ${cpuUsed}%" "警报: CPU使用率过高"
25
        writeLog "CPU usage too high"
26
    fi
27
}
28
function memUsage(){
29
    local memAvail=$(free -h | sed -n '2p' | awk '{print $NF}' | sed 's/[^0-9.]//g')
30
    local memLimit=$(jq '.memory.memory_limit' $CONF)
31
    #echo "memLimit: $memLimit"
32
    if (( $(echo "$memAvail < $memLimit" | bc -l) )); then
33
        sendEmail "可用内存不足 - 剩余可用内存为${memAvail}M" "警报: 可用内存不足"
34
        writeLog "Insufficient available memory"
35
    fi
36
}
37
function diskSpaceUsage(){
38
    local jsonArr=$(jq -r '.disk_space.mount_point[]' $CONF)
39
    IFS=$'\n' read -rd '' -a mountList <<< "$jsonArr"
40
    local diskUsageLimit=$(jq '.disk_space.disk_usage_limit' $CONF)
41
    local diskAvailLimit=$(jq '.disk_space.disk_avail_limit' $CONF)
42
    for mountPoint in ${mountList[@]}; do
43
        local diskUsage=$(df -h $mountPoint | sed -n '2p' | awk '{print $(NF-1)}' | sed 's/[^0-9.]//g')
44
        local diskAvail=$(df -h $mountPoint | sed -n '2p' | awk '{print $(NF-2)}' | sed 's/[^0-9.]//g')
45
        echo "mount point: $mountPoint; usage size: $diskUsage; avail: $diskAvail"
46
        if [ "$diskUsage" -gt "$diskUsageLimit" ] || [ "$diskAvail" -lt "$diskAvailLimit" ]; then
47
            sendEmail "挂载点\"$mountPoint\" 可用磁盘空间不足 - 剩余磁盘空间为${diskUsage}G" "警报: 可用磁盘空间不足"
48
            writeLog "Mount point: $mountPoint; Insufficient available disk space"
49
        fi
50
    done
51
}
52
function connectionStatus(){
53
    local amount=$(jq '.connection_status.amount' $CONF)
54
    local jsonArr=$(jq -r '.connection_status.status[]' $CONF)
55
    IFS=$'\n' read -rd '' -a STATUS <<< "$jsonArr"
56
    for element in ${STATUS[@]}; do
57
        #echo "element: $element"
58
        count=$(netstat -ant | grep -c ${element})
59
        if [ $count -gt $amount ]; then
60
            sendEmail "链接状态异常：${STATUS}状态链接数过多, 当前数量${count}个" "警报: 异常状态链接数过多"
61
            writeLog "Connection status exception: too many $STATUS connections"
62
        fi
63
    done
64
}
65
function containerStatus(){
66
    local containerList=$(docker ps -a | awk '{print $1}' | tail -n +2)
67
    for container in $containerList; do
68
        local status=$(docker inspect ${container} | jq -r '.[0].State.Status')
69
        if [ "$status" != "running" ]; then
70
            sendEmail "容器状态异常：${container}-${status}" "警报: 容器状态异常"
71
            writeLog "Container shut down: ${container}"
72
        fi
73
    done
74
}
75
function serviceStatus(){
76
    local jsonArr=$(jq -r '.services[]' $CONF)
77
    IFS=$'\n' read -rd '' -a SERVICE_LIST <<< "$jsonArr"
78
    for service in ${SERVICE_LIST[@]}; do
79
        #echo "service: $service"
80
        pgrep -x "$service" > /dev/null
81
        if [ $? -eq 1 ]; then
82
            sendEmail "服务状态异常: ${service} 进程关闭" "警报: 服务状态异常"
83
            writeLog "Service shut down: ${service}"
84
        fi
85
    done
86
}
87

88
while getopts "c:e:l:h" opt; do
89
    case $opt in
90
    c)
91
        CONF="$OPTARG"
92
        ;;
93
    e)
94
        EMAIL_CONF="$OPTARG"
95
        ;;
96
    l)
97
        LOGFILE="$OPTARG"
98
        ;;
99
    h)
100
        echo "Usage: ./monitor.sh -c CONFIGURATION_FILE.json -e EMAIL_CONTACT.json -l LOGFILE.log"
101
        exit 0
102
        ;;
103
    \?)
104
        exit 1
105
        ;;
106
    esac
107
done
108
if [ -z "$CONF" ] || [ -z "$EMAIL_CONF" ] || [ -z "$LOGFILE" ]; then
109
    echo "错误: 未提供正确的文件路径"
110
    echo "Try './monitor -h' for more information"
111
    exit 1
112
fi
113
if [ ! -f "$CONF" ] || [ ! -f "$EMAIL_CONF" ] || [ ! -f "$LOGFILE" ]; then
114
    echo "错误: 未找到文件"
115
    exit 1
116
else
117
    cpuUsage
118
    memUsage
119
    diskSpaceUsage
120
    connectionStatus
121
    containerStatus
122
    serviceStatus
123
fi

1
{
2
    //cpu监控项，cpu使用率大于该值时报警
3
    "cpu": {
4
        "cpu_limit": 90
5
    },
6
    //内存监控项，可用内存低于该值时报警
7
    "memory": {
8
        "memory_limit": 0.5
9
    },
10
    //磁盘监控项
11
    "disk_space": {
12
        //磁盘挂载点
13
        "mount_point": [
14
            "/"
15
        ],
16
        //磁盘使用率和空闲空间，大于指定的使用率或空闲空间小于指定的值时均报警
17
        "disk_usage_limit": 80,
18
        "disk_avail_limit": 5
19
    },
20
    //网络链接监控
21
    "connection_status": {
22
        //受监控的链接状态类型
23
        "status": [
24
            "TIME_WAIT",
25
            "SYN_SENT"
26
        ],
27
        //大于这个数量时报警
28
        "amount": 500
29
    },
30
    //那些服务受监控
31
    "services": [
32
        "nginx"
33
    ]
34
}

1
{
2
    "user1": {
3
        "email": "your@email.com",
4
        "smtp_server": "smtp.qq.com:465",
5
        "passwd": "*****",
6
        "cert": "qq.crt"
7
    },
8
    "user2": {
9
        "email": "your@email.com",
10
        "smtp_server": "smtp.qq.com:465",
11
        "passwd": "*****",
12
        "cert": "qq.crt"
13
    }
14
}