Data Lakehouse and Data Mesh: Architecture, Governance and Open Problems

Analytical data platforms have been rebuilt twice in about fifteen years. The warehouse gave way to the data lake, and the lake is now being overlaid with open table formats that add transactions to files in cloud object storage, the arrangement now called a lakehouse. Over the same period the data mesh has proposed that domain teams, not a central group, should own analytical data and publish it as products. The two are often named together. They are different kinds of thing: one is a storage and query architecture, the other an operating model for people and responsibilities. This review follows the published record on both. We trace the route from warehouse to lakehouse and set out what a published comparison of three open table formats measured. We then examine the evidence for the mesh, which rests on a synthesis of practitioner writing and a few qualitative studies. Metadata catalogues, data governance and the FAIR principles are treated as ground the two ideas share. No new experiment is reported. The open problems include the absence, in the work we read, of measured outcomes for any mesh adoption.