OMCHOKSI108/my-cloud-data-lake

🌊 Zero-Cost Cloud Data Lake on Hugging Face

A complete serverless data pipeline that converts 12GB+ of CSV data to optimized Parquet format and serves it via a FastAPI on Hugging Face Spaces.

🏗️ Architecture Overview

📁 Local CSV Data (12GB+) 
    ↓
🔄 Conversion Script (DuckDB + SNAPPY)
    ↓  
📦 Parquet Files (4.4GB - 67% compression)
    ↓
☁️ Hugging Face Dataset
    ↓
🚀 Serverless FastAPI
    ↓
🌐 Public API Endpoint

📊 Dataset Information

• Source: 793 CSV files across 6 timeframes
• Assets: 100+ financial instruments (Forex, Crypto, Stocks, Commodities)
• Timeframes: 1min, 5min, 15min, 30min, 1hr, 4hr, 1day
• Format: OHLCV (Open, High, Low, Close, Volume)
• Compression: 67% size reduction (13.4GB → 4.4GB)
• Storage: Hugging Face Datasets

🚀 Quick Start

1. API Access

• Live API: https://omchoksi108-forexdatalake.hf.space
• Documentation: https://omchoksi108-forexdatalake.hf.space/docs
• Dataset: https://huggingface.co/datasets/omchoksi108/forex-cloud-data-lake

2. API Endpoints

Endpoint	Method	Description
/	GET	API information
/health	GET	Health check
/describe	GET	Dataset schema
/preview	GET	Quick data preview
/query	POST	Execute SQL queries

3. Example Usage

Health Check

curl https://omchoksi108-forexdatalake.hf.space/health

Get Dataset Schema

curl https://omchoksi108-forexdatalake.hf.space/describe

Preview Data

curl "https://omchoksi108-forexdatalake.hf.space/preview?limit=5"

Execute SQL Query

curl -X POST https://omchoksi108-forexdatalake.hf.space/query \
  -H "Content-Type: application/json" \
  -d '{
    "sql_query": "SELECT ts, close, volume FROM data WHERE close > 100 LIMIT 10",
    "limit": 100
  }'

📁 File Structure

├── requirements_local.txt          # Local dependencies
├── scripts/
│   ├── 1_convert_to_parquet.py  # CSV → Parquet conversion
│   └── 2_upload_to_hf.py       # Upload to Hugging Face
├── api_deploy/
│   ├── main.py                  # FastAPI server
│   ├── requirements.txt          # API dependencies
│   └── Dockerfile              # Container configuration
├── parquet_data/               # Converted Parquet files
└── forexdatalake/             # Hugging Face Space files

🛠️ Local Setup

Prerequisites

• Python 3.9+
• Hugging Face account with write permissions

Installation

pip install -r requirements_local.txt

Environment Setup

Create .env file:

HF_TOKEN=your_hf_write_token_here
HF_USERNAME=omchoksi108

Data Conversion

python scripts/1_convert_to_parquet.py

Upload to Hugging Face

python scripts/2_upload_to_hf.py

🔧 Technical Details

Data Processing

• Engine: DuckDB (in-memory analytical database)
• Compression: SNAPPY (balanced speed/size)
• Format: Apache Parquet (columnar storage)
• Memory Usage: Optimized for large datasets

API Features

• Serverless: No server management required
• Streaming: Direct Parquet queries without download
• Caching: HTTP file caching for performance
• Security: SQL injection protection
• Monitoring: Health checks and error handling

Performance Metrics

• Conversion Rate: ~1.8 files/second
• Compression Ratio: 67% size reduction
• Query Response: Sub-second for most queries
• Concurrent Users: Handles multiple simultaneous requests

📈 Data Schema

Each Parquet file contains:

ts        TIMESTAMP    -- Data timestamp (YYYY-MM-DD HH:MM)
open      DOUBLE       -- Opening price
high      DOUBLE       -- Highest price
low       DOUBLE       -- Lowest price
close     DOUBLE       -- Closing price
volume    BIGINT       -- Trading volume

🔒 Security Features

• SQL Validation: Only SELECT statements allowed
• Query Limits: Configurable result limits
• Input Sanitization: Prevents SQL injection
• Private Dataset: Access controlled via Hugging Face
• Token Security: No hardcoded credentials

🚀 Deployment

Automatic Deployment

1. Push changes to forexdatalake folder
2. Hugging Face automatically builds Docker container
3. API becomes available at your Space URL

Manual Deployment

cd forexdatalake
git add .
git commit -m "Update API"
git push origin master

📊 Query Examples

Filter by Symbol

SELECT * FROM data 
WHERE filename LIKE '%BTCUSD%' 
LIMIT 100

Time Range Analysis

SELECT ts, close, volume 
FROM data 
WHERE ts BETWEEN '2023-01-01' AND '2023-12-31'
  AND symbol = 'EURUSD'
ORDER BY ts DESC

Price Statistics

SELECT 
    AVG(close) as avg_close,
    MIN(close) as min_close,
    MAX(close) as max_close,
    COUNT(*) as total_records
FROM data 
WHERE symbol = 'BTCUSD'

Volume Analysis

SELECT 
    DATE(ts) as date,
    SUM(volume) as daily_volume,
    AVG(close) as avg_price
FROM data 
WHERE symbol = 'ETHUSD'
GROUP BY DATE(ts)
ORDER BY date DESC
LIMIT 30

🔄 Data Updates

Adding New Data

1. Add CSV files to appropriate timeframe folders
2. Run conversion script
3. Upload updated Parquet files
4. API automatically queries latest data

Schema Changes

• Update conversion script for new columns
• Re-run conversion process
• Upload updated dataset

🐛 Troubleshooting

Common Issues

Token Permissions

Error: 403 Forbidden - You don't have rights to create dataset
Solution: Create new HF token with write permissions

Memory Issues

Error: Out of memory during conversion
Solution: Reduce DuckDB memory_limit in script

API Timeouts

Error: Request timeout
Solution: Add LIMIT clause to queries

Performance Optimization

• Use specific column selection instead of SELECT *
• Add appropriate WHERE clauses
• Use LIMIT for large result sets
• Consider time-based partitioning for queries

📚 Resources

• DuckDB Documentation: https://duckdb.org/docs/
• FastAPI Documentation: https://fastapi.tiangolo.com/
• Hugging Face Datasets: https://huggingface.co/docs/datasets/
• Parquet Format: https://parquet.apache.org/

🤝 Contributing

1. Fork the repository
2. Create feature branch
3. Make changes
4. Test thoroughly
5. Submit pull request

📄 License

This project is licensed under the MIT License.

🙏 Acknowledgments

• DuckDB Team - High-performance analytical database
• Hugging Face - Free hosting and infrastructure
• FastAPI - Modern web framework
• Parquet Community - Efficient columnar storage

---

🌟 Star this project if you find it useful!

📧 Contact: For issues and questions, use the Issues tab on Hugging Face.