Raft agreement

For strong ⼀ Consensus algorithm , At present, the most used is Raft agreement .

Raft Implementation cases of mature protocols

• Ant gold JRaft;
• Zookeeper Of ZAB;
• Consul Of Raft;
• Baidu braft;
• Apache Of Ratis;

Raft role

• follower （Follower）： Ordinary people , Silently receiving and receiving messages from leaders , When the leader's heartbeat message times out , I'll take the initiative to stand up , Recommend yourself as a candidate .
• The candidate （Candidate）： Candidates will ask other nodes to vote RPC news , Inform other nodes to vote , If you win a majority of the votes , Just be promoted to be a leader .
• The leader （Leader）： Handle write requests 、 Manage log replication and constantly send heartbeat information （ Maintain leadership ）.

Random timeout characteristics

Raft The algorithm has the characteristics of random timeout , The timeout interval of each node waiting for the heartbeat information of the leader node is random . To prevent multiple nodes from voting at the same time , The election timeout allocated by each node is also random .

The election process

1. The initial state , All nodes are followers .
2. When there is no leader in the cluster , The node with the smallest waiting timeout will not wait for the leader's heartbeat , Timeout occurred .
3. Overtime self recommendation , At this point, the node will add its own tenure number , And recommend yourself as a candidate , Vote for yourself first , Then send a request to vote to other nodes RPC news , Ask them to elect themselves leaders .
4. The election vote , Other nodes receive the candidate's RPC When the news , And numbered 1 Term of office , Have you ever voted for , Then vote for the candidate , Then add your term number .
5. New leaders emerge , The candidate won a majority of the votes during the election timeout , Then it will be the new leader of this term .
6. The normal state , Leaders will periodically send heartbeat messages to other services .