Questions about implementation

[JoanFM]

Hello team,

I have been reading a lot through the documentation and a little about the raft consensus protocol.

I think I am getting most of the concepts and abstractions but I still have some questions.

• The first question is about this diagram https://github.com/hashicorp/raft/blob/main/docs/apply.md

Is there a mechanism in raft or in this raft implementation that can handle cases where one of the ApplyEntries or FSM apply, or FSMSnapshot fails in one replica but not in the others? Or is the assumption that all replicas behave completely equivalent and functionally there is no way that one works and not the others? If this is handled, what is the mechanism to achieve so?

• Another question is how to manage the starting of the different peers in the cluster?

Is there always a need for an External client to register the Peers to all the rest? Should it come with a detailed configuration file which is updated when peers come and go? Is there a comprehensive documentation about this part of the algorithm and how to set things up?

• The last question is somehow related to the last one, and is about whether or not there is an example of an application using this library running on Kubernetes with potentially some StatefulSets to see how the bootstrapping may look like?

Thank you very much,