raft_kvs

• benchmark raft replication log performance
• test how long does follower switch leader

some code refer from go.etcd.io/etcd/contrib/raftexample

Modify

• only leader node can post data
• add isLeader api
• add getSelfState api
• more code description
• fix don't rm old snapshot bug
• fix don't rm old wals bug
• fix restore data raise block bug in snapshot
• format code
• support leader judge
• modify http api
• add benchmark
• more...

Test

config dep

export GOPATH=<directory>
cd <directory>/src/go.etcd.io/etcd/contrib/raftexample
go build -o raftexample

install goreman

go get github.com/mattn/goreman

run raft cluster

make run

run benchmark

make batch1w
make check

make batch10w
make check

make batch20w
make check

make batch100w
make check

Method

show leader and followrer

make leader

check a kv record, check value in all nodes

make check

show raft nodes info

make info

Benchmark

benchmark raft log replication performance

benchmark method

• In a leader node, active make 100w record, and sync data to followers.
• leader and followers all in a node.

8 cpu core, 16 mem

Result

Concurrent 1 workers

3 w QPS/S

Concurrent 20 workers

4 w QPS/S

Concurrent 100 workers

4.5 w QPS/S

Concurrent 500 workers

4 w QPS/S