About This Talk
At IT’s Tinkoff 2022, I gave a candid, experience-based account of the Data Mesh journey at Yandex Go — covering not just the theory, but the real-world consequences of decoupling a monolithic data warehouse into domain-owned data products. The title is intentionally provocative: Data Mesh can break your DWH if you’re not prepared.
The Problem
Yandex Go’s centralized data platform reached its scaling limits. Request queues grew longer. Priority conflicts became constant. Domain-specific knowledge was lost in translation between business teams and the central data team. Data Mesh seemed like the answer — give data ownership to domain teams who understand the data best.
Key Ideas
Why We Chose Data Mesh — The centralized model was at its limits. Wait times for new data products were unacceptable. Ownership was unclear. Quality suffered because the central team couldn’t be experts in every domain. Data Mesh’s promise of domain ownership aligned with our organizational pain.
The Decoupling Process — We identified domain boundaries, separated shared tables into domain-owned products, established contracts between domains, and built self-serve infrastructure so domain teams could operate independently. This was organizational surgery, not just a technical migration.
What Broke — Cross-domain queries became exponentially more complex. Performance degraded when queries had to span multiple domain-owned datasets. Governance gaps appeared — who ensures naming conventions? Who resolves conflicts? Team readiness varied dramatically; some domains thrived while others struggled.
What Worked — Domain teams that embraced ownership became faster and more autonomous. Data quality improved because the people who understood the data were now responsible for it. The feedback loop between data producers and consumers shortened dramatically.
Why It Matters
Data Mesh is not a silver bullet. This talk provides a realistic, warts-and-all account of what happens when you apply the paradigm at scale. The lessons are valuable whether you’re considering Data Mesh or just want to understand the trade-offs of centralized vs. decentralized data architectures.
