raft-kv

About this project

Distributed storage services are the backbone of modern Internet applications. However, building a robust distributed storage service from scratch is challenging, particularly in orchestrating the subtle cases in distributed consensus.

The Raft-KV addresses this by providing an abstract Raft consensus layer that can be integrated with any key-value (KV) storage engine. This separation of concerns between the storage layer and the distributed consensus layer enhances modularity, making it easier and more efficient for developers to build reliable storage clusters.

The project's goal is to enable developers to build robust distributed storage clusters by simply "plugging" their custom KV storage engine into the Raft consensus layer. This provides a flexible and reliable solution for distributed data management.

this project serves as a comprehensive study of distributed systems, consensus algorithms, and modern C++ development.

Getting started

Build

mkdir -p raft-kv/build
cd raft-kv/build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j8

Running a cluster

First install goreman, which manages Procfile-based applications.

goreman start

Test

install redis-cli, a redis console client.

redis-cli -p 63791
127.0.0.1:63791> set mykey myvalue
OK
127.0.0.1:63791> get mykey
"myvalue"

remove a node and replace the myvalue with "new-value" to check cluster availability:

goreman run stop node2
redis-cli -p 63791
127.0.0.1:63791> set mykey new-value
OK

bring the node back up and verify it recovers with the updated value "new-value":

redis-cli -p 63792
127.0.0.1:63792> KEYS *
1) "mykey"
127.0.0.1:63792> get mykey
"new-value"

benchmark

redis-benchmark -t set,get -n 100000 -p 63791

====== SET ======
  100000 requests completed in 1.35 seconds
  50 parallel clients
  3 bytes payload
  keep alive: 1

96.64% <= 1 milliseconds
99.15% <= 2 milliseconds
99.90% <= 3 milliseconds
100.00% <= 3 milliseconds
73909.83 requests per second

====== GET ======
  100000 requests completed in 0.95 seconds
  50 parallel clients
  3 bytes payload
  keep alive: 1

99.95% <= 4 milliseconds
100.00% <= 4 milliseconds
105485.23 requests per second