Four Years of Data Mesh in Practice
summary
Four years of practice yields concrete claims: a reference implementation that deploys a data product in ~25–30 minutes using a JSON spec, Open Policy Agent checks, per‑domain AWS accounts, S3, EMR/Spark and Airflow or Databricks on Azure, and a catalog implemented with Apache Atlas/Azure Purview. Results: governance remains the hardest problem, executive sponsorship is required, small wins scale when repeated, and some clients get useful data in months (one in two months, another in three weeks) though a pure data mesh rarely appears.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 3:56What data mesh is
- 8:54Domain ownership
- 13:51Data platform basics
- 18:07Four years of lessons
- 21:36Executive sponsorship
- 23:02Data as product
- 26:29DevOps foundations
- 29:17Scaling data mesh
- 31:04Governance challenges
- 33:52Self-service platforms
- 36:47Multicloud cost control
- 37:51Cross-company data sharing
- 39:14Spreadsheets and literacy gaps
- 40:18Mesh vs fabric vs Lakehouse
- 41:01Closing and Q&A
