etcd: etcd go client fails when querying a cluster with a down node

Describe the bug The etcd go client fails if multiple https endpoints are specified when the client is initialised and the first etcd endpoint is unavailable.

To Reproduce I setup an etcd cluster with http (port 2378) and https (port 2379) listeners. Then used the etcd go client library to query the cluster. Then I took down the first unit listed when the client was established (in my case the client on port 10.53.82.119). The http go client continues to work but the https one fails.

https client:

	ctx, _ := context.WithTimeout(context.Background(), requestTimeout)

	cfg := clientv3.Config{
		Endpoints:   []string{"https://10.53.82.119:2379" ,"https://10.53.82.150:2379", "https://10.53.82.157:2379"},
		DialTimeout: 5 * time.Second,
	}

	cert := "/home/liam/tls_vault_certs/etcd-cert.pem"
	key := "/home/liam/tls_vault_certs/etcd.key"
	ca := "/home/liam/tls_vault_certs/etcd-ca.pem"
	tls := transport.TLSInfo{
		TrustedCAFile: ca,
		CertFile:      cert,
		KeyFile:       key,
	}

	tlscfg, err := tls.ClientConfig()
	cfg.TLS = tlscfg

	cli, err := clientv3.New(cfg)
	if err != nil {
		log.Fatal(err)
	}
	defer cli.Close()
	kv := clientv3.NewKV(cli)

Fails with: 2018-07-21 11:34:05.728613 I | context deadline exceeded

http client:

        ctx, _ := context.WithTimeout(context.Background(), requestTimeout)

        cfg := clientv3.Config{
                Endpoints:   []string{"http://10.53.82.119:2378" ,"http://10.53.82.150:2378", "http://10.53.82.157:2378"},
                DialTimeout: 5 * time.Second,
        }

        cli, err := clientv3.New(cfg)
        if err != nil {
                log.Fatal(err)
        }
        defer cli.Close()
        kv := clientv3.NewKV(cli)