5 ms·
The Raft Consensus Algorithm
- pavpanchekha 12y agoHave the authors seen [ http://en.wikipedia.org/wiki/Paxos_(computer_science) http://en.wikipedia.org/wiki/Paxos_(computer_science) ](PAXOS)? It seems related.
- freefrag 12y agoYes. They even reference it on their website: "Raft is a consensus algorithm that is designed to be easy to understand. It's equivalent to Paxos in fault-tolerance and performance. The difference is that it's decomposed into relatively independent subproblems, and it cleanly addresses all major pieces needed for practical systems."
- jbp 12y agoPAXOS is very complicated, Raft is direct response to it by simplifying the consensus algorithm. From the article: "It's equivalent to Paxos in fault-tolerance and performance. The difference is that it's decomposed into relatively independent subproblems, and it cleanly addresses all major pieces needed for practical systems." One of the authors of Raft is Prof. John Ousterhout(TCL creator, among other things)
- krat0sprakhar 12y agoOne of the most lucid explanations of the Raft Consensus Algorithm that I've ever seen is the one in John Ousterhout and Diego Ongaro's Raft User Study[0]. The 2 videos do an excellent job of explaining both the Paxos and Raft algorithms and the key differences between them. John's narration is extremely clear and the slides are very informative. Definitely recommended! Raft is a consensus algorithm designed to be easier to understand than Paxos. To measure Raft's understandability, we conducted an experimental study using CS students at two universities. We recorded a video lecture of Raft and another of Paxos, and created corresponding quizzes. This page makes our materials available for anyone interested. We think these are valuable resources for anyone learning consensus (whether Raft or Paxos or both). [0] - https://ramcloud.stanford.edu/~ongaro/userstudy/ https://ramcloud.stanford.edu/~ongaro/userstudy/ Raft Video - https://www.youtube.com/watch?v=YbZ3zDzDnrw https://www.youtube.com/watch?v=YbZ3zDzDnrw Paxos Video - http://www.youtube.com/watch?v=JEpsBg0AO6o&feature=youtu.be http://www.youtube.com/watch?v=JEpsBg0AO6o&feature=youtu.be
- icedog 12y agoDirectly to the results: https://ramcloud.stanford.edu/~ongaro/userstudy/graphs.pdf https://ramcloud.stanford.edu/~ongaro/userstudy/graphs.pdf
- penguindev 12y agoThis is pretty awesome and I think only real hardcore folks who need time-guaranteed leader failover (like, I don't know, a heart pacemaker or something) would stick with paxos. (Assuming paxos can provide that).
- penguindev 12y agohmm, actually watching the paxos video posted above - even basic paxos needs randomized delays to avoid livelock with multiple proposers. ouch. so why would people choose it over raft?
- tedsuo 12y agoThe only issue I have with raft (besides bugs in the various implementations I have to deal with at work) is that it is not yet proven to truly be simpler than Paxos. In other words, is it a genuinely simpler algorithm that provides the the same guarantees as Paxos, or is it simpler because it leaves edge cases unspecified, forcing implementors to grind through the edge cases with heuristics? If it's the second, it's not actually simpler, it's just worse. I'm not saying that it isn't genuinely simpler. Just that in discussions about raft, I see most people blindly assuming that it is. But I don't believe that it has been proven yet. I think this is the nuclear power plant part of the bikeshed discussion. [1] If you scrub to the 50 minute mark of this talk[2] by Leslie Lamport (creator of Paxos), my coworker David Varvel asks him this question and he gives more or less the same answer. [1] http://en.wikipedia.org/wiki/Parkinson's_law_of_triviality http://en.wikipedia.org/wiki/Parkinson's_law_of_triviality [2] http://channel9.msdn.com/Events/Build/2014/3-642 http://channel9.msdn.com/Events/Build/2014/3-642
- justinsb 12y agoIs there a good open-source Paxos implementation? There are many open-source Raft implementations, some of which are good. goraft powers etcd, for example. I believe most developers just want to consume a working library. It seems likely that the plethora of Raft libraries and dearth of Paxos libraries is because it is easier to write a Raft library, but I also don't think it matters _why_, it only matters that there are good Raft libraries.
- tedsuo 12y agoCoreOs (etcd team) is currently rewriting their raft implementation because it had a lot of problems. I don't believe there is a "good" implementation out there yet, because it's all very new. And should an implantation be "good", AKA implement the spec without introducing incidental errors, will it still have edge cases because the spec has edge cases? This is all currently unknown. It's precisely these assumptions that I'm pointing out. A quick google search brings up libpaxos and other open paxos implementations, so it appears that they exist.
- wcummings 12y agoThe RAFT paper is incredibly comprehensible, and clearly written with an audience of engineers in mind (it even specs out signatures for RPC calls). Definitely worth a read, if you have't read it already. I saw a good talk about RAFT given by a Basho engineer, video and slides are here: http://www.meetup.com/Erlang-NYC/events/131394712/ http://www.meetup.com/Erlang-NYC/events/131394712/ The speakers Erlang implementation: https://github.com/andrewjstone/rafter https://github.com/andrewjstone/rafter
- brunoqc 12y agoCan the Raft Consensus works with only two nodes?
- antirez 12y agoTechnically speaking, it should work, however it is no longer available during partitions since the only 2 nodes partition creates two minority partitions, so the algorithm can't promote a new leader, nor make progresses since new log entries can't reach the majority. Basically you have the same properties using a single node without consensus, if not for the data redundancy. However it is not something like: one node dies, the other can continue. TLDR: 3 nodes is the practical minimum.