Hi, > DNS request is an additional cost in nodes communication.
I do not think we will abandon using IP. Either using IP or hostname depends on users. If they are sure they have a good DNS, they can use DNS; If they like IP or the nodes do not set the DNS, they can use IP. If they like, they can also mix-use the hostname and IP addresses. We do not need to care..... > We should make the relationship between node IP, node hostname and node > identifier be clear. IMO, node identifier is the only option. Best, ----------------------------------- Xiangdong Huang School of Software, Tsinghua University 黄向东 清华大学 软件学院 Eric Pai <[email protected]> 于2021年8月26日周四 下午4:46写道： > > Hi, Xiangdong, > > Supporting hostname is really a good feature! However, we should consider > these cons carefully: > > 1. DNS request is an additional cost in nodes communication. In cloud > environment, the latency may be as large as hundreds of milliseconds(e.g. > across available regions), and local DNS cache is also an issue in some > extreme scenarios. > 2. We should make the relationship between node IP, node hostname and node > identifier be clear. A node should not be treated as a new one if it changes > its IP or hostname. Or allowing to change IP only. > > 在 2021/8/25 下午11:15，“Xiangdong Huang”<[email protected]> 写入: > > Hi, > > Today Julian and I discussed some logic of the cluster module. > Both of us feel it strange that we replace the hostname to IP before a > cluster node starts up. > I think it is not a very common way. > > hostname sometimes is very very helpful. For example, in K8S > environment, a node's hostname never changes, but its IP may change. > > Best, > ----------------------------------- > Xiangdong Huang > School of Software, Tsinghua University > > 黄向东 > 清华大学 软件学院 >
