Data Lake — Architecture for Storing Raw Data

Data lake is a central repository for raw data in any format. From structured tables to unstructured logs — everything on cheap object storage.

What is Data Lake¶

Stores data in raw form — schema-on-read.

Architecture¶

• Storage — S3, GCS, ADLS
• Formats — Parquet, Avro, JSON
• Catalog — Glue, Hive Metastore
• Compute — Spark, Trino, DuckDB

s3://data-lake/
├── raw/           # Bronze
│   ├── orders/
│   └── events/
├── processed/     # Silver
│   └── orders/
├── curated/       # Gold
│   └── daily_revenue/
└── _metadata/

What to Avoid (Data Swamp)¶

• Missing catalog
• No governance
• Small files — thousands of 1KB files
• Missing lineage

Summary¶

Data lake with Table Formats and governance becomes lakehouse — reliable foundation for analytics.