🥱 Semi-toy implenetation of RAFT consensus protocol.

Just for fun

Note: refactor wip

Note: Depreprecated

• Frontend - HTTP
• Backend - intracluster RPC calls via gRPC
• Fault tolerant load balancing using HAProxy
• PostgreSQL as persistent storage (had big troubles with simple text files)

Note: Depreprecated

1. Set environmetal variables:

• PG_HOST
• PG_USER
• PG_PASS
• PG_DB
• PG_PORT

2. go run main.go <id> <ids> inside node/cmd directory.
   • <id> stands for current node id, so it should be unique. Ids stand for node number in whole cluster, this number should be equal in all node's setups.
   • <ids> should be odd. Why? Consider case when we have just 2 nodes. Each voted for another and both think it is a Leader (Pure brain split, I've observed 2 nodes sending heartbeats and accepting it from each other). We cannot make Quorum = 2 (because candidate is not set votedFor for itself). Maybe force turning to Follower inside HeartBeat RPC? It might work, but unstable and I am not sure, because it can lead to extra election iterations. This problem is the same for each even n.
3. curl -L --post301 -i -d 'command=woopwoop' http://localhost:6070/replicate

• Raft