一般情况无法知道etcd是否正常，需要自行监控

var (
    dbcli      *clientv3.Client
    WatchClose bool
)

var etcdDialConfig = clientv3.Config{
    Endpoints:            []string{IP + ":2379"},
    DialTimeout:          5 * time.Second,
    DialKeepAliveTime:    5 * time.Second,
    DialKeepAliveTimeout: 5 * time.Second,
    DialOptions:          []grpc.DialOption{grpc.WithBlock()},
}

func newClient(config clientv3.Config) error {
    rlogger.FuncEntry(types.ModuleOamCm, nil)
    db, err := clientv3.New(config)
    if err != nil {
        rlogger.Trace(types.ModuleOamCm, rlogger.ERROR, nil, "init etcd fail with", err)
    } else {
        dbcli = db
        rlogger.Trace(types.ModuleOamCm, rlogger.INFO, nil, "init etcd success")
        WatchClose = true
    }
    return err
}

func healthCheck() {
    for {
        rlogger.FuncEntry(types.ModuleOamCm, nil)
        time.Sleep(2 * time.Second)
        //_, err := dbcli.Dial(configure.CmSysConf.EtcdSerAddr + ":2379")
        timeoutCtx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
        defer cancel()
        _, err := dbcli.Status(timeoutCtx, etcdDialConfig.Endpoints[0])
        if err != nil {
            for {
                err = newClient(etcdDialConfig)
                if err == nil {
                    break
                }
            }
        }
    }
}

需要注意，for循环读取的watch不会更新，改为如下(当重新加载etcd后通过dbapi.WatchClose告诉协程EtcdWatch)：

func EtcdWatch() {
WATCH:
    dbapi.WatchClose = false
    watch := dbapi.DbWatchPrefix(key)
    for {
        select {
        case wresp, ok := <-watch:
            if ok {
                for _, ev := range wresp.Events {
                                //todo some
                }
            }
        default:
            //使用ticker会阻塞,必须加延时
            time.Sleep(1 * time.Second)
            if dbapi.WatchClose {
                goto WATCH
            }
        }
    }
}