Raft Consensus

Roadmap Feature

Raft consensus is planned for a future release. This page describes the intended architecture.

TensorDB’s clustering layer will use the Raft consensus protocol for leader election and log replication.

Architecture

┌─────────┐    ┌─────────┐    ┌─────────┐
│ Node 1  │◄──►│ Node 2  │◄──►│ Node 3  │
│ (Leader) │    │(Follower)│    │(Follower)│
└─────────┘    └─────────┘    └─────────┘

Design Goals

• Strong consistency: All reads and writes go through the Raft leader
• Automatic failover: If the leader fails, followers elect a new leader
• Append-only replication: TensorDB’s immutable fact model maps naturally to Raft’s replicated log
• Bitemporal preservation: All temporal metadata is replicated faithfully

How It Will Work

1. Leader election: Raft elects a leader node among cluster members
2. Write path: Writes go to the leader, which replicates to followers via the Raft log
3. Commit: Once a majority acknowledge, the write is committed
4. Read path: Reads can be served from any node (with configurable consistency)

Consistency Levels

Level	Description	Latency
Strong	Read from leader only	Higher
Bounded staleness	Read from follower within time bound	Medium
Eventual	Read from any node	Lowest