• 中文文档

Raft-Simple

• A simple raft implementation。Only support key(string) value(string)

Server

Efficiency Optimization

• Send leader's Log on batch, act on 100 message or 10ms no more new message
• heartbeat, at most wait for 100ms, avoid to elect cause by timeout
• async on message receiving and message replicating
• Leader send messages in parallel
• Avoid to message disorder， using pipeline processing

Communication between Nodes

• Communication by GRPC between nodes
• Heartbeat or Log replication
• Election

Persistence

• Only storage Log Index and Log Content
• file1：
  • Header RAFT01 "RAFT"+2 byte version No
  • Log count storage by uint64
  • Log Index, uint64 uint32, uint64 uint32，means kv index address and content length
• file2：
  • storage kv Log Content, connected with file1

Log Replication

• Leader send message to Follower by heartbeat, Follower save Log
• Leader receives responses from most Followers, update state machine on local, then response client a success
• Follower commit the Log to state machine when it realized this Log was committed by Leader

Timeout

• Heartbeat Tiimeout： 150-300ms，Follower will translate to Candidate, if the heartbeat was not received in time
• Election Timeout: 150-300ms, Candidate will add the term and launch an election again, if Candidate didn't receive more than half of the votes

Node Expansion

• single node expansion(support later), test on 3 nodes now
• Log full copy and replay(support later)

Client