Raft Consensus on Microcontrollers

in progress

Distributed consensus protocol running on a 5-node ESP32-S3 cluster communicating over ESP-NOW. Leader election, log replication, and fault tolerance – on real hardware. Kill any 2 nodes and the cluster keeps running. Bring them back and they catch up automatically.

Architecture

+------------------+     ESP-NOW (250B frames, 1-10ms latency)     +------------------+
|   Node 0         |<---------------------------------------------->|   Node 1         |
|   ESP32-S3       |         broadcast heartbeats                   |   ESP32-S3       |
|   BME280 (I2C)   |         unicast vote responses                 |   BME280 (I2C)   |
|   WS2812 LED     |         unicast AppendEntries replies          |   WS2812 LED     |
|   OLED SSD1306   |                                                |   OLED SSD1306   |
|   SPI FRAM       |                                                |   SPI FRAM       |
+------------------+                                                +------------------+

Every node is identical hardware and firmware. Node ID is set via NVS config at flash time. 5 nodes total in full mesh.

Raft Protocol

Each node is always Follower, Candidate, or Leader. Two core RPCs:

RequestVote – a candidate solicits votes after its election timeout fires. A node votes for at most one candidate per term, and only if the candidate’s log is at least as up-to-date.

AppendEntries – the leader replicates log entries and sends heartbeats (empty entries). Followers reject if their log doesn’t match at prevLogIndex.

Persistent State

Stored in SPI FRAM (MB85RC256V) for unlimited write endurance:

Field	Size	Notes
`currentTerm`	4 bytes	Latest term seen
`votedFor`	1 byte	Vote in current term (0xFF = none)
`logEntryCount`	4 bytes	Number of committed entries

Log entries stored in SPI Flash (W25Q32) with circular buffer and wear-leveling.

Timing

Parameter	Value	Rationale
Heartbeat	200 ms	~20x ESP-NOW RTT
Election timeout	1000-2000 ms	Randomized to prevent split votes
RPC timeout	100 ms
Snapshot threshold	100 entries	Triggers InstallSnapshot for lagging nodes

Application Layer

The replicated state machine drives three features:

Sensor fusion – median temperature/humidity from BME280 across all nodes, with outlier rejection
Synchronized LEDs – WS2812B RGB state replicated via the Raft log, all nodes display the same color
Key-value store – serial console CLI with SET/GET/GETL commands, plus cluster STATUS and fault injection (KILL/REVIVE)

Status

ESP-NOW transport and node identity complete
Leader election and heartbeats in progress
Log replication next