A minimal implementation of Raft in Go. Demo video.
This blog post walks through the code.
NOT FOR PRODUCTION USE.
Things that are missing:
- Client serial identifier for each message (maybe the most obvious missing function/actual bug)
- Search for "session" in Diego Ongaro's thesis for more details.
- Snapshotting/checkpointing and associated state transfer
- Configuration change protocol
- Rigged up to Jepsen
- Surely much else
Not particularly aggressive yet but does basic correctness and stress testing.
Read more about the details of what it tests in the README.
$ cd cmd/stress
$ go run main.go util.go
With the go-deadlock
package turned off and the default sync
package on, I get throughput of around 20k-40k entries/second with
this stress test.
Try out the builtin distributed key-value API.
$ cd cmd/kvapi && go build
$ rm *.dat
$ ./kvapi --node 1 --http :2021 --cluster "0,:3030;1,:3031;2,:3032"
$ ./kvapi --node 1 --http :2021 --cluster "0,:3030;1,:3031;2,:3032"
$ ./kvapi --node 2 --http :2021 --cluster "0,:3030;1,:3031;2,:3032"
Remember that requests will go through the leader (except for if we
turn that off in the /get
request). So you'll have to try sending a
message to each server until you find the leader.
To set a key:
$ curl http://localhost:2020/set?key=y&value=hello
To get a key:
$ curl http://localhost:2020/get\?key\=y
- In Search of an Understandable Consensus Algorithm: The Raft paper.
- raft.tla: Diego Ongaro's TLA+ spec for Raft.
- Jon Gjengset's Students' Guide to Raft
- Jack Vanlightly's Detecting Bugs in Data Infrastructure using Formal Methods (TLA+ Series Part 1): An intro to TLA+.
And useful implementations I looked at for inspiration and clarity.
- Hashicorp's Raft implementation in Go: Although it's often quite complicated to learn from since it actually is intended for production.
- Eli Bendersky's Raft implementation in Go: Although it gets confusing because it uses negative numbers for terms whereas the paper does not.
- Jing Yang's Raft implementation in Rust.