💡 Inspiration

In high-stakes distributed systems—such as high-voltage electrical grid monitoring, automated trading risk gateways, and emergency dispatch networks—sudden traffic storms and hardware outages cause catastrophic bufferbloat and death spirals.

When an unexpected 10x surge hits a traditional backend, FIFO queues swell, clients trigger retry storms, and critical emergency shutoff commands ($P_0$ Breaker Trips) get starved behind millions of trivial heartbeat telemetry packets.

We built ObsidioCore to answer the Track 2: Obsidio (The Siege) and Track 1: Fundamentum challenges: to build a distributed telemetry engine that doesn't just survive contact with extreme load, but guarantees strict sub-5ms latency SLAs for life-critical operations while gracefully degrading non-critical traffic under 10,000+ RPS bursts.


⚡ What It Does

ObsidioCore is an ultra-resilient, distributed high-throughput event processing and risk mitigation engine engineered with:

  • Lock-Free Circular Ring Buffer (LMAX Disruptor Pattern): Zero-copy power-of-two bitmask indexing for microsecond event ingestion.
  • CoDel Adaptive Priority Load Shedding ($L = \lambda W$): Monitors queue dwell time dynamically. Under congestion, it progressively sheds low-priority telemetry ($P_3 \to P_2$) while ensuring $P_0$ critical signals experience 0% shed rate and sub-5ms processing SLA.
  • Self-Healing 3-State Circuit Breaker: Sliding-window error evaluator with automated trip to OPEN, exponential backoff with randomized jitter, and concurrency-isolated HALF-OPEN recovery probing.
  • Adaptive Token Bucket & Leaky Bucket Smoothers: In-memory token replenishers with dynamic downstream backpressure scaling.
  • Bayesian Gaussian Process Regression (GPR): Live $\pm 1.96\sigma$ (95% Bayesian Confidence Ribbon) to predict and visualize queue dwell-time drift and bufferbloat risks.
  • Glassmorphic Real-Time Command Console: Live Chart.js latency percentile streams ($p50, p95, p99$), throughput velocity bars, Chaos Fault Injector, and an audible ✨ 1-Click Judge Tour (Web Speech API).

🛠️ How We Built It

  • Backend & Concurrency: FastAPI, AsyncIO, Python 3.11, Uvicorn ASGI Server.
  • Core Systems Architecture:
    • Lock-free bitwise modulo ring buffer (core/ring_buffer.py)
    • Adaptive token replenishment limiter (core/rate_limiter.py)
    • 3-state self-healing circuit breaker with sliding window math (core/circuit_breaker.py)
    • CoDel dwell-time priority shedder (core/load_shedder.py)
    • Real-time fault injection chaos engine (core/chaos_engine.py)
  • Real-time WebSockets & Telemetry: Asynchronous pub/sub aggregator calculating rolling NumPy percentiles ($p50, p95, p99, \text{max}$).
  • Frontend: HTML5, Tailwind CSS, Lucide Icons, Chart.js, Vanilla ES6 WebSockets with automatic reconnection.
  • Testing & Benchmarks: 20/20 Pytest suite (100% passing) and multi-threaded siege stress generator.

📊 Measured Benchmark Results

During automated siege testing (2,000 concurrent requests at 4,000+ RPS burst):

Metric Measured Value SLA Target Status
P0 Critical p99 Latency 2.85 ms $< 50.0\text{ ms}$ PASSED (100% On-Time)
Median p50 Latency 1.14 ms $< 10.0\text{ ms}$ PASSED
Ingestion Peak Velocity 5,420+ RPS $> 3,000\text{ RPS}$ PASSED
System Crash / 500 Rate 0.00% $0.00\%$ ZERO CRASHES
P0 Critical Delivery Rate 100.0% $100.0\%$ PERFECT RETENTION
Automated Test Pass Rate 100% (20/20 Passing) $100\%$ VERIFIED

🧠 What We Learned

  1. Bufferbloat is deadlier than CPU saturation: Queuing delays caused by deep FIFO buffers cause cascading client timeouts long before CPU cores are pinned. Active Queue Management (CoDel) based on dwell time is essential.
  2. Jitter prevents thundering herds: Without randomized jitter in circuit breaker HALF-OPEN recovery probes, all queued clients retry simultaneously, immediately re-tripping the breaker.
  3. Power-of-two bitmasking: Replacing modulo % capacity with & (capacity - 1) in the circular buffer yielded massive performance improvements during tight loop event ingestion.

🚀 Try It Out (Quickstart)

# 1. Clone repository
git clone https://github.com/Prateek312413/Catalyst-2026.git
cd Catalyst-2026

# 2. Install dependencies
pip install -r requirements.txt

# 3. Launch Mission Control Console (Auto-opens browser to http://localhost:8000)
python run.py

# 4. Run Pytest Verification Suite (20/20 Passing)
pytest tests/ -v

# 5. Run Standalone Siege Benchmark
python tests/stress_benchmark.py 1000 30

Built With

Share this project:

Updates

Submission history