Yup! I dislike how it is being sold as some thing new.. it is a Paxos/Viewstamp-Replication step by step guide, if you want to replicate a log where there is a strong causal relation between updates. If you want to replicate entities that are independent there many simplifications that can be made for failover.

The good thing that has come out of this is that in internal meetings with folks who have read the Google Paxos paper (not my fav paper, leads to blind pessimism about the prospects of implementing paxos without understanding how to contextualize) you can still sell the same solution by calling it raft.

Paxos was published after the first randomized consensus algorithms, and Paxos has been shown to be a variation on those principles, so not sure the Paxos worship is all that warranted..

(Please read my bio before responding..)

Certainly Paxos is not the be-all-end-all of randomized consensus algorithms. However, what algorithm do you think we should worship?

Hmmm... Ben Ors randomized consensus algorithm? Can't beat the simplicity of that one...

Yes! In fact published before Paxos, and studiously ignored by Lamport + shown to be equivalent.

yea raft is basically a multi-paxos variant.

It's much simpler though.

Multipaxos was designed with "okay, we have paxos, now let's extend it". Raft was designed with all of the requirements in it from the start.