Feature Request: add Raft::client_current_leader API

[xu-cheng]

In addition to the existing Raft::client_read API, may I suggest to add an Raft::client_current_leader API like the following:

pub async fn client_current_leader(&self) -> Result<NodeId, ClientCurrentLeaderError> {
    match self.client_read().await {
        Ok(()) => Ok(cur_node_id),
        Err(ClientReadError::ForwardToLeader(Some(id))) => Ok(id),
        Err(ClientReadError::ForwardToLeader(None)) => Err(ClientCurrentLeaderError::LeaderUnknown),
        Err(ClientReadError::RaftError(e)) => Err(ClientCurrentLeaderError::RaftError(e)),
    }
}

This would be helpful to certain applications, where the leader is required to do certain preprocessing before submitting the real Raft write request.

Also, it may be a good idea to have another similar API, which returns the current leader without sending any Raft requests for performance reason. Something like:

pub async fn client_current_presumptive_leader(&self) -> Result<NodeId, ClientCurrentLeaderError> {
    if self.is_leader() {
        Ok(cur_node_id)
    } else if let Some(leader_id) = self.leader_id {
        Ok(leader_id)
    } else {
        Err(ClientCurrentLeaderError::LeaderUnknown)
    }
}

When using this API, it is up to the application's responsibility to check who is the real leader later.

[thedodd]

I do think that this is a great idea. For the latter, I will often just use the Raft's metrics handle, which has a field indicating if the node is currently the leader or not. We could probably build upon that mechanism for the second item you've mentioned.

[xu-cheng]

Thanks for the advice. I created PR #99.

I only implemented for the second API mentioned above. Because, I found that the first API may not be that useful. At the same time, I am not sure the first API can actually return the confirmed leader if it is called from a non-leader node.