# Data Model: Raft Core Replication

## Entities

- **Peer**
  - Fields: `id (u64)`, `region_id (u64)`, `state (Leader/Follower/Candidate)`, `term (u64)`, `commit_index (u64)`, `last_applied (u64)`
  - Relationships: owns `RaftStorage`; exchanges `RaftLogEntry` with other peers.
  - Constraints: single region scope for this phase; fixed voter set of 3.

- **RaftLogEntry**
  - Fields: `index (u64)`, `term (u64)`, `command (bytes)`, `context (bytes, optional)`
  - Relationships: persisted in `raft_log` CF; applied to state machine when committed.
  - Constraints: indices strictly increasing; term monotonic per election; applied in order.

- **HardState**
  - Fields: `current_term (u64)`, `voted_for (u64)`, `commit_index (u64)`
  - Relationships: persisted in `raft_state` CF; loaded at startup before participating.
  - Constraints: must be flushed atomically with log appends when advancing commit index.

- **ConfState**
  - Fields: `voters (Vec<u64>)`
  - Relationships: persisted in `raft_state` CF; defines quorum (majority of 3).
  - Constraints: static for this phase; changes require future joint consensus.

- **ReplicationState**
  - Fields: `match_index (u64)`, `next_index (u64)`, `pending (bool)`
  - Relationships: maintained per follower in memory; not persisted.
  - Constraints: drives AppendEntries backoff and progress.

## State Transitions

- Peer transitions: Follower → Candidate → Leader on election; Leader → Follower on higher term or failed election.
- Log application: when `commit_index` advances, apply entries in order to state machine; `last_applied` increases monotonically.
- Recovery: on restart, load `HardState`, `ConfState`, and log; reconcile with leader via AppendEntries (truncate/append) before applying new entries.