Client support for etcd dynamic cluster membership reconfiguration

When running etcd server on the cloud (AWS, for example), as instances get terminated new instances with different IP addresses are launched to replace them. With time, an etcd cluster that used to be reachable through a set of IP addresses, will have a totally different set of addresses.

In order for the client to be able to keep connected and working with an ever changing set of cluster IP addresses, somethings are required. It seems that this client already has some of them. It already supports the configuration of both URIResolver and LoadBalancer which are essential for properly discovering all the nodes of a cluster but also implementing fallback strategies, as described in etcd client architecture docs https://etcd.readthedocs.io/en/latest/client-architecture.html#client-architecture

Nevertheless, I was not able to achieve the behaviour described in such architecture docs. The problems are the following:

1. once an etcd host goes down, the client does not fallback to another host
2. when a new host is added, it’s not automatically added in the channel

How do you think would be the correct way to achieve 1 and 2?

Ideally it would be possible to “fix” a channel by managing connections to hosts, adding and removing connections.

[rcillo]

@User979269852 I think this is part of grpc-core. I have a line like this in my project import io.grpc.LoadBalancerRegistry;

[luomengY]

@User979269852 I think this is part of grpc-core. I have a line like this in my project import io.grpc.LoadBalancerRegistry;

How is the “DnsSrvUriResolver” defined? Is there any specific implementation code?