监控告警管理脚本示例

#!/bin/bash
# monitor_alert_management.sh
# from:www.itpux.com.qq113257174.wx:itpux-com
# web: http://www.fgedu.net.cn

# 定义变量
PROMETHEUS_HOME="/usr/local/prometheus"
GRAFANA_HOME="/usr/local/grafana"
ALERTMANAGER_HOME="/usr/local/prometheus"

# 检查Prometheus状态
check_prometheus() {
    echo "检查Prometheus状态..."
    curl -s http://localhost:9090/metrics > /dev/null 2>&1
    if [ $? -eq 0 ]; then
        echo "Prometheus运行正常"
        return 0
    else
        echo "Prometheus运行异常"
        return 1
    fi
}

# 检查Grafana状态
check_grafana() {
    echo "检查Grafana状态..."
    curl -s http://localhost:3000/api/health > /dev/null 2>&1
    if [ $? -eq 0 ]; then
        echo "Grafana运行正常"
        return 0
    else
        echo "Grafana运行异常"
        return 1
    fi
}

# 检查Alertmanager状态
check_alertmanager() {
    echo "检查Alertmanager状态..."
    curl -s http://localhost:9093/metrics > /dev/null 2>&1
    if [ $? -eq 0 ]; then
        echo "Alertmanager运行正常"
        return 0
    else
        echo "Alertmanager运行异常"
        return 1
    fi
}

# 重启监控服务
restart_services() {
    echo "重启监控服务..."
    
    # 重启Prometheus
    pkill -f prometheus
    sleep 2
    cd $PROMETHEUS_HOME && ./prometheus --config.file=prometheus.yml &
    echo "Prometheus已重启"
    
    # 重启Grafana
    pkill -f grafana-server
    sleep 2
    cd $GRAFANA_HOME && ./bin/grafana-server &
    echo "Grafana已重启"
    
    # 重启Alertmanager
    pkill -f alertmanager
    sleep 2
    cd $ALERTMANAGER_HOME && ./alertmanager --config.file=alertmanager.yml &
    echo "Alertmanager已重启"
}

# 检查监控目标
check_targets() {
    echo "检查监控目标..."
    curl -s http://localhost:9090/api/v1/targets | jq '.data.activeTargets[] | {target: .labels.instance, state: .state}'
}

# 主流程
echo "=== 监控告警平台管理开始 ==="

# 检查服务状态
check_prometheus
check_grafana
check_alertmanager

# 检查监控目标
check_targets

# 询问是否重启服务
read -p "是否重启监控服务？(y/n): " answer
if [ "$answer" = "y" ]; then
    restart_services
fi

echo "=== 监控告警平台管理完成 ==="