[SPIKE] High-Throughput Trade-Matching Engine via Redis Streams & WebSockets

[jotel-dev]

[SPIKE] Architecture & Benchmark Spike: Low-Latency Trade-Matching Engine via Redis Streams

🏷️ Metadata

• Type: Spike / Architecture & Benchmarking
• Difficulty: Hard
• Estimated Time (with AI aid): ~60 minutes
• Primary Stack: TypeScript, Node.js, Redis Streams, WebSockets

---

📘 Background & Context

When buyers create cash requests, liquidity providers in the same geographic region compete to fulfill them. Relying solely on database polling introduces latency. We need to evaluate an in-memory, event-driven order matching architecture using Redis Streams.

---

🎯 Problem Statement & Scope

Evaluate and prototype an order-matching pipeline for cash requests:

1. Stream Producer: Prototype publishing cash requests to Redis Stream stream:cash_requests.
2. Consumer Group Matcher: Implement a worker node using Redis Consumer Groups to match cash requests to registered providers within a 5km radius under 10ms processing time.
3. Benchmarking: Benchmark throughput (operations/sec) under simulated loads of 1,000 concurrent requests.
4. Deliverables:
   • Architectural document: docs/spikes/redis-matching-engine.md.
   • Runnable benchmark script: examples/matching_engine_benchmark.ts.

---

🤖 AI-Assisted Implementation Guide (1-Hour Workflow)

1. Step 1 (15m): Ask AI to write a TypeScript Redis Stream producer and consumer group worker module using ioredis.
2. Step 2 (25m): Have AI write a stress-test generator spawning 1,000 concurrent cash request payloads and measuring latency distribution (p50, p95, p99).
3. Step 3 (20m): Instruct AI to compile latency benchmarks and architectural trade-offs into docs/spikes/redis-matching-engine.md.

---

✅ Acceptance Criteria

• Published benchmark report in docs/spikes/redis-matching-engine.md.
• Runnable benchmark script outputting throughput metrics.
• Demonstrated p95 processing latency < 15ms.