portfolio project · distributed systems

ActiveMQ + EMQX
Distributed Math System

A production-grade distributed system that routes arithmetic computation requests across worker nodes via dual message brokers, with real-time observability through the ELK stack and results persisted to PostgreSQL.

View on GitHub See it running ↓

3,070results in postgres

99.52%success rate

1,213msgs enqueued/dequeued

6MQTT clients connected

OVERVIEW

What this project demonstrates

Built to showcase real-world distributed systems design, this project implements a horizontally scalable computation pipeline. Math operations submitted by a load generator are serialised as messages and distributed across worker nodes through two complementary brokers — ActiveMQ for JMS-style enterprise queuing and EMQX for lightweight MQTT pub/sub.

Completed results are persisted to PostgreSQL and every event is streamed to Elasticsearch for indexing. A custom Kibana dashboard surfaces operation duration, success rate, worker utilisation, and a live duration timeline — demonstrating end-to-end production observability.

ARCHITECTURE

System design

Two views of the system — the overall conceptual diagram showing all components and message flow, and the deployment architecture showing how services are distributed across the 5 VMs.

OVERALL SYSTEM DIAGRAM

Overall system architecture showing all 5 servers, message flow, and components

↗ click to enlarge

Full system overview — 5 servers, dual brokers, ELK observability

S1 client · S2 ActiveMQ · S3 coordinator (EMQX, Kafka, PostgreSQL, ELK) · S4 Worker A · S5 Worker B

DEPLOYMENT ARCHITECTURE

TECH STACK

Components

MESSAGE BROKER · JMS

Apache ActiveMQ

Handles math.requests and math.results queues with durable delivery. 1,213 messages enqueued and fully dequeued with zero pending.

MESSAGE BROKER · MQTT

EMQX

6 clients connected — mqtt-amq-bridge, math-worker-a/b, kafka-producer, dispatcher-mqtt, db-writer — all showing Connected status.

SEARCH & INDEXING

Elasticsearch

Indexes 2,306+ calculation events in real time. Powers the Kibana Math System Dashboard with per-operation and per-worker breakdowns.

OBSERVABILITY

Kibana

Custom dashboard showing avg duration per operation, 99.52% success rate donut, recent calculations table, and a 4-hour duration timeline.

PERSISTENCE

PostgreSQL

Stores all completed results. A live query confirms 3,070 rows in the results table — verifying end-to-end message delivery.

LOAD GENERATION

Python scripts

load-generator.py dispatches randomised operations (add, subtract, multiply, divide, sqrt, power, factorial) alternating between worker_a and worker_b.

HOW IT WORKS

Message lifecycle

Load generator dispatches tasks

load-generator.py publishes randomised math operations every 10 seconds to the ActiveMQ math.requests queue, including the operation type and operands.

ActiveMQ queues and routes

ActiveMQ holds messages durably and delivers them to the next available consumer. The DLQ monitor confirms zero dead-letter messages — every task is picked up.

EMQX bridges via MQTT

The mqtt-amq-bridge republishes tasks over MQTT. EMQX distributes them to subscribed workers (math-worker-a, math-worker-b) using lightweight pub/sub.

Workers compute and publish results

Stateless workers execute the operation in sub-millisecond time, then publish the result back to math.results. Results alternate cleanly between worker_a and worker_b.

Results persisted to PostgreSQL

The db-writer MQTT client consumes results and inserts them into PostgreSQL. A live SELECT COUNT(*) confirms 3,070 rows — matching the number of tasks dispatched.

ELK stack provides observability

Logstash ships all events to Elasticsearch. Kibana's Math System Dashboard shows 99.52% success rate, per-operation duration breakdowns, and a live 5-minute-interval timeline.

SCREENSHOTS