rkv

Reinventing the wheel - raft based KV store using Go & gRPC (protobuf)

Intro

A distributed key value store based on Raft consensus algorithm. It supports:

1. Raft election
2. Replication & log shipping
3. Log compaction & snapshots
4. Consensus based writes, with batching to improve throughput
5. Auto follower to leader proxy for write operations (set/del)
6. gRPC based client/server & node/node communication
7. Abstracted raft layer which can potentially be used with other statemachines, or different communication protocols

Build

cd cmd/rkv
go build .
cd cmd/rkvclient
go build .

This builds two executables, one in each folder: rkv (rkv server), rkvclient (rkv client)

Run locally

Start raft server nodes.

Below cmds start a 3 node cluster on loopback. To watch logs easily, it's better to start each node in a different terminal.

./rkv -nodeid 0 -addresses localhost:27015,localhost:27016,localhost:27017
./rkv -nodeid 1 -addresses localhost:27015,localhost:27016,localhost:27017
./rkv -nodeid 2 -addresses localhost:27015,localhost:27016,localhost:27017

Run client against any nodes for set/get/del

./rkvclient set -address localhost:27015 -key somekey0 -value v0
./rkvclient set -address localhost:27016 -key sk1 -value v1
./rkvclient set -address localhost:27016 -key sk2 -value v2
./rkvclient get -address localhost:27016 -key somekey0
./rkvclient get -address localhost:27017 -key sk1
./rkvclient del -address localhost:27015 -key sk2

Benchmark

Below benchmark was run against the leader node directly:

./rkvclient benchmark -address localhost:27016 -times 20000
=====================================================
Benchmark with 1 connection, 100 concurrent clients:
SET
  Total set clients   : 100
  Total set calls     : 20000
  Total set success   : 20000
  Total time taken    : 989 ms
  Average set latency : 4 ms
GET
  Total get calls     : 20000
  Total get success   : 20000
  Total time taken    : 511ms
  Average get latency : 2ms

Happy coding. Peace.

MIT © sidecus