Distributed Sensor System

A lightweight, dynamic distributed sensor network supporting automatic leader election via the bully algorithm and ordered group messaging using a simple sequencer.

Overview

• Dynamic host discovery via UDP broadcast over BROADCAST_PORT (default: 9999).
• Peers maintain a groupView and orderedPeerList for membership and ordering.
• The leader acts as both group coordinator and sequencer for ordered multicast.
• Failure detection uses leader heartbeats and member ACKs.
• Mocked sensor data generation and replication across peers via totally ordered multicast.

Leader Election (Bully Algorithm)

• Priority: peers compare by peer_id (UUID) lexicographically. This yields a deterministic total order without needing host or port tie-breakers.
• Trigger: election starts when members detect leader heartbeat timeout or when a leader failure is processed.
• Flow:
  1. A peer sends ELECTION:<sender_id> to all peers with higher priority.
  2. Higher peers respond OK and start their own election.
  3. If no OK is received within election_timeout, the initiator announces COORDINATOR:<peer_id> and becomes leader.
  4. All peers update leader_id; the leader starts heartbeats.

Message Types

• Discovery: NEW_PEER_REQUEST (UDP), ADDED TO NETWORK (TCP).
• Membership change: VIEW_CHANGE:<orderedPeerList>-<groupView>-<view_id> (TCP).
• Heartbeats: HEARTBEAT / HEARTBEAT_ACK (TCP).
• Election: ELECTION:<peer_id>, OK, COORDINATOR:<peer_id> (TCP).
• Sequencer: SEQUENCER_REQUEST:<payload> (TCP).
• Sensor data: DATA|sensor_id|timestamp|temperature|humidity|pressure (payload inside sequencer flow).

Failure Detection

• Leader sends HEARTBEAT to members every heartbeat_interval; expects HEARTBEAT_ACK within heartbeat_timeout.
• On timeout for a member, the leader removes the member from the view and multicasts VIEW_CHANGE.
• Members run leader_check_thread and start bully election if the leader heartbeat is not observed within heartbeat_timeout.

Configuration Knobs

• heartbeat_interval (default: 5s)
• heartbeat_timeout (default: 20s)
• election_timeout (default: 5s)
• sensor_interval_seconds (default: 15s per peer)

Run

From the project root:

# Start three peers
python main_multithreading.py

# Fail-stop test (kills a non-leader and checks membership updates)
python test_fault_tolerance.py

# Bully leader election test (detects current leader, kills it, verifies new leader)
python test_leader_election.py

Logs

• Tests redirect per-peer output to logs/{peer_id}_log.txt (e.g., logs/peer2_log.txt).
• Look for Received COORDINATOR announcement: <peer> or became the Group Leader via election in these log files to verify election.

Data Replication

• Each peer periodically generates mocked measurements (temperature, humidity, pressure) every 15s.
• Measurements are sent to the leader (sequencer) and broadcast with a total order to all peers.
• Every peer appends delivered measurements to per-sensor CSVs under its own directory: data/{peer_id}/{sensor_id}.csv.
• This means each peer holds a full replicated set of all sensors' data, easing recovery after failures.
• CSV columns: sequence,timestamp,sensor_id,temperature_c,humidity_pct,pressure_hpa.

Run-to-run reset

• On peer startup, its data/{peer_id} directory is cleared and recreated to avoid appending to prior runs.
• The test test_data_handling.py also clears the base data/ directory before launching peers for a clean test environment.

Windows Notes

• You may see warnings about SO_REUSEPORT not being supported.
• Multiple broadcast listeners across processes can cause bind conflicts on the same port. The system still functions using whichever listener successfully binds.

Extending Priority

• To change election priority, adjust get_priority() in Peer.py to use other criteria (e.g., numeric IDs or timestamps).