Storage: databases, warehouses, lakes and lakehouses

Where data lives decides what it costs and what you can do with it.

Store	What it is	Good for	Examples
Operational database	The app's OLTP database	Running the product	PostgreSQL, MySQL
Data warehouse	Managed OLAP database for structured tables	SQL analytics, dashboards	BigQuery, Snowflake, Redshift
Data lake	Files in cheap object storage, any format	Raw data, huge volumes, ML, logs	Amazon S3, Google Cloud Storage, Azure Data Lake
Lakehouse	A lake plus table features: transactions, schemas, time travel	One copy of data for both SQL and ML	Delta Lake, Apache Iceberg, Apache Hudi

The history in one line: warehouses were reliable but expensive and only for tables; lakes were cheap and flexible but easily turned into a "data swamp"; lakehouses add warehouse-style guarantees on top of lake storage.

File formats

Format	Layout	Use it for
CSV	Row, plain text	Exchanging small files with people and vendors
JSON	Row, nested text	APIs and events
Parquet	Column, compressed, typed	Analytics on a lake: the default choice
Avro	Row, compact, carries its schema	Streaming messages, Kafka

Partitioning

Big tables are split into partitions, usually by date, so a query for one day reads one folder instead of all of history:

text

 s3://company-lake/orders/order_date=2026-10-16/part-0001.parquet

 s3://company-lake/orders/order_date=2026-10-17/part-0001.parquet

on the job Many cloud warehouses charge by the amount of data a query scans. Columnar formats, partitioning and selecting only the columns you need (never SELECT * on a big table) directly cut the bill.