flanneld检查脚本

k8s集群node节点flanneld服务down的检查重启脚本

# cat check_flanneld.sh 
#!/bin/bash

# 服务名称
FLANNEL_SERVICE="flanneld"
DOCKER_SERVICE="docker"

# 检查 flanneld 是否运行
if ! systemctl is-active --quiet $FLANNEL_SERVICE; then
    echo "$(date '+%Y-%m-%d %H:%M:%S') [WARN] $FLANNEL_SERVICE 已停止，开始重启..."

    # 重启 flanneld
    systemctl restart $FLANNEL_SERVICE

    # 等待几秒确保 flanneld启动完成
    sleep 5

    echo "$(date '+%Y-%m-%d %H:%M:%S') [INFO] $FLANNEL_SERVICE 已重启"

    # 重启 docker
    echo "$(date '+%Y-%m-%d %H:%M:%S') [INFO] 开始重启 $DOCKER_SERVICE..."
    systemctl restart $DOCKER_SERVICE
    echo "$(date '+%Y-%m-%d %H:%M:%S') [INFO] $DOCKER_SERVICE 已重启"
else
    echo "$(date '+%Y-%m-%d %H:%M:%S') [INFO] $FLANNEL_SERVICE 正常运行"
fi

设置定时任务执行

# crontab -e
*/30 * * * * /bin/bash /data/scripts/check_flanneld.sh  >> /var/log/check_flanneld.log 2>&1