Consensus - ljcom/operadb GitHub Wiki
🧭 Consensus in OperaDB
OperaDB uses an event-driven architecture, and in its current form, runs as a single-node system. However, for multi-node or distributed deployments, a lightweight consensus mechanism becomes important to ensure consistency and trust across nodes.
🧠 Philosophy of Consensus in OperaDB
OperaDB treats events as the source of truth. All state is derived by replaying events using deterministic reducers. Therefore, the key requirement for consensus is:
- Ensuring event order and content are identical across all nodes.
- Ensuring all reducers compute the same state from the same ordered events.
⚙️ Proposed Consensus Workflow
✅ Event-Hash Agreement
-
Each node receives an event and computes:
eventHash = hash(prevEventHash + currentEvent) -
Nodes broadcast their computed hash to peers.
-
If 2/3 of nodes agree on the hash:
- The event is committed to the chain
- Reducers run and state is updated
-
If disagreement occurs:
- Event is rejected or delayed
- Nodes may re-sync or trigger rollback
🌀 Alternative: Epoch-Based Quorum (Optional)
OperaDB may implement an epoch + voting system in the future:
- Events are grouped into batches (epochs)
- Nodes vote to commit or reject batches
- Majority (e.g., 2/3) required to reach consensus
🔐 Key Principles
| Concept | Description |
|---|---|
| Deterministic Reducers | All nodes must produce the same output from the same input |
| Event Hashing | Events are linked via cryptographic hashes |
| No Global Clock | Consensus uses event ordering, not timestamps |
| Quorum Agreement | Decisions are based on 2/3 agreement |
📌 Comparison vs. Blockchain Consensus
| Feature | OperaDB | Blockchain (e.g., Ethereum) |
|---|---|---|
| Consensus Type | Lightweight hash-voting | Proof of Work / Stake |
| Target Use Case | Business workflow DB | Trustless global ledger |
| Reducer Output | Deterministic state | Irreversible block state |
| Performance | High (single writer) | Lower (due to mining/fees) |
🚧 Current Status
| Feature | Status |
|---|---|
| Single-node replay | ✅ Implemented |
| Multi-node replay | ⚠️ Planned |
| Hash-based voting | 🔄 Experimental |
| Byzantine Tolerance | ❌ Not supported |
OperaDB aims to remain lightweight but optionally support distributed consensus in clustered or decentralized environments.