nsqd: SUB: fix ephemeral channel race condition with reconnecting cli…

…ent.

It's possible for a client reconnecting quickly and subscribed to an ephemeral channel to race with nsqd's cleanup of said ephemeral channel, as documented in nsqio/go-nsq#206.

Fixes nsqio#883

nsqd/protocol_v2.go

@@ -616,7 +616,15 @@ func (p *protocolV2) SUB(client *clientV2, params [][]byte) ([]byte, error) {
 	}
 
 	topic := p.ctx.nsqd.GetTopic(topicName)
+	for topic.Exiting() {
+		time.Sleep(1 * time.Microsecond)
+		topic = p.ctx.nsqd.GetTopic(topicName)
+	}
 	channel := topic.GetChannel(channelName)
+	for channel.Exiting() {
+		time.Sleep(1 * time.Microsecond)
+		channel = topic.GetChannel(channelName)
+	}
 	channel.AddClient(client.ID, client)
 
 	atomic.StoreInt32(&client.State, stateSubscribed)