Hi, chris +1 for identifying nodes solely by nodeid
Best ---------------- Xinyu Tan On 2024/07/24 14:33:20 Christofer Dutz wrote: > Hi all, > > I’m currently working on https://github.com/apache/iotdb/pull/12914 > > Here one of the problems was, that in a cluster with 3 datanodes, where the > dn_rpc_address is set to 0.0.0.0, the command “show datanodes” lists each > node with an address of 0.0.0.0. So, if someone wants to remove a data-node > by its IP, the cli will not find the corresponding node as it thinks it’s > 0.0.0.0. However, if you use the cli to delete 0.0.0.0, then it deletes all > nodes. > > Now my initial fix for this, was, that if a data-node registers and says his > dn_rpc_address is 0.0.0.0, that instead of this, the IP from which we are > getting the request is being used (Obviously this one exists and belongs to > the data-node registering). > The problem is that this usually will be the dn_internal_address instead of > the dn_rpc_address. > > Now we could use that instead, but I think it would reduce the usefulness of > the “show datanodes” command, because it could be in a cluster-internal > network, that the client can’t connect to. > So, if a client wants to know which other data-nodes there are in order to > connect to another one, this might not be helpful. > However, 0.0.0.0 is also not helpful, as I see no chance to be able to > connect to a data-node using 0.0.0.0:6667 as that’s not really a real Ip > address. > > So, I think, that possibly instead of maintaining 0.0.0.0 we should replace > this with the list of public IP addresses the data-node possesses. In this > case show data-nodes would no longer display only one IP-Address, but a list > of IP-Addresses. > > When removing a data-node (or sending any other commands to it) we could now > identify a particular data-node as only one will have the IP+Port combination. > > What do you think? > > Chris >
