Data Architecture Guide

Data Lakehouse Architecture

The lakehouse unifies data lakes and warehouses: cheap object storage with ACID transactions, schema enforcement, and fast SQL. Delta Lake, Iceberg, and Hudi make it production-ready in 2026.

Diagram Your LakehouseMedallion Architecture
18 min readFor Data Engineers & Architects

See it as a diagram

Everything below, as a diagram you can edit. Describe yours and see it in seconds.

77/2000• 3 credits left
Try:

No account needed · Editable canvas, not a picture

Why Lakehouse?

Data lakes are cheap but lack reliability. Warehouses are reliable but expensive and vendor-locked. The data lakehouse bridges both: open table formats on object storage provide warehouse features at lake prices.

Data Lake

Cheap storage, all data types, but no ACID, no schema. "Data swamp" risk.

Data Warehouse

Fast SQL, ACID, governance, but expensive per TB and vendor-locked.

Data Lakehouse

Lake pricing + warehouse reliability. Open formats, multi-engine access.

Open Table Formats: The Core Technology

FormatBest ForBacked ByEngines
Delta LakeDatabricks ecosystemsDatabricksSpark, Trino, Flink
Apache IcebergMulti-engine, open governanceApple, Netflix, AWSSpark, Trino, Flink, Dremio, StarRocks
Apache HudiUpsert-heavy, CDC workloadsUber, AWSSpark, Trino, Flink, Presto

2026 Recommendation

Apache Iceberg has the most momentum: widest engine support, best catalog interop (REST Catalog), and adoption by Snowflake, Databricks, AWS, and Google. If starting fresh, Iceberg is the safest choice.

Lakehouse Architecture Layers

Storage Layer

Cloud object storage (S3, GCS, ABFS). Parquet files organized by table format metadata. Cheap, durable, scalable.

Table Format Layer

Delta Lake / Iceberg / Hudi metadata on top of Parquet. Provides ACID transactions, time travel, schema evolution, and partition evolution.

Compute Layer

Multiple engines read the same data: Spark for ETL, Trino for ad-hoc SQL, Flink for streaming, Python for ML. No data copying.

Catalog & Governance

Unity Catalog, AWS Glue Catalog, or REST Catalog manages table metadata, access policies, and lineage across engines.

Consumption Layer

BI tools (Looker, Tableau), ML frameworks (MLflow), and APIs consume data directly from the lakehouse.

Lakehouse vs Warehouse Decision

Choose Lakehouse when:

  • Petabyte+ data at lowest storage cost
  • Unstructured data (logs, images, embeddings)
  • Multi-engine access (SQL + Spark + ML)
  • Avoid vendor lock-in with open formats

Choose Warehouse when:

  • Primarily SQL analytics workloads
  • Structured data under 100TB
  • Want fully managed experience
  • Fast time-to-value (no infra management)

FAQ

What is a data lakehouse?
A data architecture combining lake storage (cheap, open) with warehouse features (ACID, schema, SQL). Uses open table formats (Delta Lake, Iceberg, Hudi) on cloud object storage.
Delta Lake vs Iceberg vs Hudi?
Delta Lake for Databricks ecosystems. Iceberg for multi-engine and open governance (fastest growing). Hudi for upsert-heavy CDC workloads.
Lakehouse vs data warehouse?
Lakehouse for petabyte scale, unstructured data, multi-engine access, and vendor independence. Warehouse for fast SQL analytics, managed experience, and structured data under 100TB.

Diagram your lakehouse architecture

AI-generated diagrams with cloud icons for S3, Iceberg, Spark, and more. Map your entire data platform.

Try Datadef Free