IsNew is too precise

[jcjones]

In client.IsNew, we check to see that cluster info gives back a very specific set of data:

attache/src/redis/client/client.go

Lines 27 to 36 in 0db713b

	var infoMatchingNewNodes = clusterInfo{"fail", 0, 0, 0, 0, 1, 0, 0, 0, 0, 0}
	clusterInfo, err := h.GetClusterInfo()
	if err != nil {
	return false, err
	} else if *clusterInfo == infoMatchingNewNodes {
	return true, nil
	} else {
	return false, nil
	}
	}

In production, sometimes there are cluster stats messages sent and received, which causes Attache to assume a new cluster member is not actually new.

The stats messages received should be ignored. In a functioning cluster, cluster members will send pings to this new node as it comes online to assess its status. That shouldn't cause Attache to declare it kaput.

10.1.12.83:20001> cluster info
cluster_state:fail
cluster_slots_assigned:0
cluster_slots_ok:0
cluster_slots_pfail:0
cluster_slots_fail:0
cluster_known_nodes:1
cluster_size:0
cluster_current_epoch:0
cluster_my_epoch:0
cluster_stats_messages_pong_sent:2
cluster_stats_messages_sent:2
cluster_stats_messages_ping_received:2
cluster_stats_messages_received:2

10.1.12.85:20001> cluster info
cluster_state:fail
cluster_slots_assigned:0
cluster_slots_ok:0
cluster_slots_pfail:0
cluster_slots_fail:0
cluster_known_nodes:1
cluster_size:0
cluster_current_epoch:0
cluster_my_epoch:0
cluster_stats_messages_pong_sent:6
cluster_stats_messages_sent:6
cluster_stats_messages_ping_received:6
cluster_stats_messages_received:6