Skip to content

QUICK_START.md

Latest commit

 

History

History
176 lines (132 loc) · 4.4 KB

QUICK_START.md

File metadata and controls

176 lines (132 loc) · 4.4 KB

Tachyon Argus - Quick Start Guide

Get the predictive monitoring system running in 5 minutes.

Prerequisites

  • Python 3.10+
  • CUDA-capable GPU (recommended) or CPU
  • 8GB RAM minimum

Step 1: Clone and Setup Environment

# Clone the repository
git clone https://github.com/your-org/MonitoringPrediction.git
cd MonitoringPrediction

# Create conda environment
conda env create -f environment.yml
conda activate tachyon

Step 2: Generate API Key

cd Argus

# Windows
bin\setup_api_key.bat

# Linux/Mac
./bin/setup_api_key.sh

This creates .tachyon_key with your API key.

Step 3: Start the System

Option A: Full System (Inference + Demo Data + Dashboard)

# Windows
start_all.bat

# Linux/Mac
./start_all.sh

Option B: Inference Daemon Only

python src/daemons/tft_inference_daemon.py --port 8000

Step 4: Verify It's Running

# Health check (no auth required)
curl http://localhost:8000/health

# Expected response:
# {"status": "healthy", "service": "tft_inference_daemon", "running": true}

Step 5: Feed Some Data

curl -X POST http://localhost:8000/feed/data \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{
    "records": [{
      "timestamp": "2025-01-15T10:00:00",
      "server_name": "testserver001",
      "status": "healthy",
      "cpu_user_pct": 25.0,
      "cpu_sys_pct": 5.0,
      "cpu_iowait_pct": 2.0,
      "cpu_idle_pct": 68.0,
      "java_cpu_pct": 10.0,
      "mem_used_pct": 60.0,
      "swap_used_pct": 0.0,
      "disk_usage_pct": 40.0,
      "net_in_mb_s": 5.0,
      "net_out_mb_s": 3.0,
      "back_close_wait": 0,
      "front_close_wait": 0,
      "load_average": 1.5,
      "uptime_days": 30
    }]
  }'

Step 6: Get Predictions

After feeding ~100 records (warmup period):

curl -H "X-API-Key: YOUR_API_KEY" http://localhost:8000/predictions/current

Step 7: Check Fleet Health (Cascading Failure Detection)

The system automatically detects cascading failures across your server fleet:

# Get fleet health score
curl -H "X-API-Key: YOUR_API_KEY" http://localhost:8000/cascade/health

# Response shows fleet-wide status:
# {
#   "health_score": 85.2,
#   "status": "healthy",
#   "correlation_score": 0.234,
#   "anomaly_rate": 0.05,
#   "cascade_risk": "low"
# }

Step 8: Monitor Model Drift

Check if the model needs retraining:

curl -H "X-API-Key: YOUR_API_KEY" http://localhost:8000/drift/status

# Response:
# {
#   "drift_detected": false,
#   "auto_retrain_enabled": true,
#   "last_retrain": "2025-01-15T10:30:00"
# }

Access Points

Service URL Description
API http://localhost:8000 REST API for predictions
Dashboard http://localhost:8501 Web dashboard (if started)
Health http://localhost:8000/health Health check endpoint
Cascade Status http://localhost:8000/cascade/status Cascading failure detection
Fleet Health http://localhost:8000/cascade/health Fleet-wide health score
Drift Status http://localhost:8000/drift/status Model drift monitoring

Key Features

  • Multi-Target Prediction: Predicts CPU, memory, swap, I/O wait, and load simultaneously
  • Cascading Failure Detection: Identifies fleet-wide issues even when individual servers appear healthy
  • Automatic Model Drift Detection: Monitors prediction accuracy and triggers retraining when needed
  • Continuous Learning: Model improves automatically as drift is detected

What's Next?

Troubleshooting

"Model not found" error

  • You need a trained model in Argus/models/
  • Either download a pre-trained model or train one (see Training Guide)

"insufficient_data" error

  • The daemon needs ~100 data points before predictions work
  • Keep feeding data or use the demo generator

Connection refused

  • Check the daemon is running: curl http://localhost:8000/health
  • Check the port isn't blocked by firewall

GPU not detected

  • Verify CUDA is installed: python -c "import torch; print(torch.cuda.is_available())"
  • The system will fall back to CPU if no GPU is available