This is a series of projects on a key-value storage system built with the Raft consensus algorithm. These projects are inspired by the famous MIT 6.824 course but aim to be closer to industry implementations. The whole course is pruned from TiKV and re-written in Go. After completing this course, you will have the knowledge to implement a horizontally scalable, highly available, key-value storage service with distributed transaction support and a better understanding of TiKV implementation.
The whole project is a skeleton code for a kv server and a scheduler server at the beginning, and you need to finish the core logic step by step:
- Project1: build a standalone key-value server
- Project2: build a highly available key-value server with Raft
- Project3: support multi Raft group and balance scheduling on top of Project2
- Project4: support distributed transaction on top of Project3
Important note: This course is still under development, and the documentation is incomplete. Any feedback and contribution is greatly appreciated. Please see help wanted issues if you want to join in the development.
Here is a reading list for the knowledge of distributed storage system. Though not all of them are highly related with this course, they can help you construct the knowledge system in this field.
Also, you’d better read the overview of TiKV and PD's design to get a general impression on what you will build:
- TiKV
- https://pingcap.com/blog-cn/tidb-internal-1/ (Chinese Version)
- https://pingcap.com/blog/2017-07-11-tidbinternal1/ (English Version)
- PD
- https://pingcap.com/blog-cn/tidb-internal-3/ (Chinese Version)
- https://pingcap.com/blog/2017-07-20-tidbinternal3/ (English Version)
First, please clone the repository with git to get the source code of the project.
git clone https://github.com/pingcap-incubator/tinykv.gitThen make sure you have go >= 1.13 toolchains installed. You should also have make installed.
Now you can run make to check that everything is working as expected. You should see it runs successfully.
Similar to the architecture of TiDB + TiKV + PD that separates the storage and computation, TinyKV only focuses on the storage layer of a distributed database system. If you are also interested in the SQL layer, please see TinySQL. Besides that, there is a component called TinyScheduler acting as a center control of the whole TinyKV cluster, which collects information from the heartbeats of TinyKV. After that, the TinyScheduler can generate some scheduling tasks and distribute them to the TinyKV instances. All of them are communicated via RPC.
The whole project is organized into the following directories:
kv: implementation of the TinyKV key/value store.proto: all communication between nodes and processes uses Protocol Buffers over gRPC. This package contains the protocol definitions used by TinyKV, and the generated Go code that you can use.raft: implementation of the Raft distributed consensus algorithm, which is used in TinyKV.scheduler: implementation of the TinyScheduler which is responsible for managing TinyKV nodes and generating timestamps.log: log utility to output log based on level.
Please follow the course material to learn the background knowledge and finish code step by step.
After you finish the whole implementation, it becomes runnable. You can try TinyKV by deploying it onto a real cluster, and interact with it through TinySQL.
make
It builds the binary of tinykv-server and tinyscheduler-server to bin dir.
Put the binary of tinyscheduler-server, tinykv-server and tinysql-server into a single dir.
Under the binary dir, run the following commands:
mkdir -p data
./tinyscheduler-server
./tinykv-server -path=data
./tinysql-server --store=tikv --path="127.0.0.1:2379"
mysql -u root -h 127.0.0.1 -P 4000