Data Engineering with Medallion Architecture (Miki Eto) (z-library.sk, 1lib.sk, z-lib.sk)

【One-Line Pitch】 A production-focused guide to building and safely operating bronze/silver/gold medallion data platforms, for data engineers and architects who must ship pipelines that survive audits, schema changes, and 2 AM incidents. 【Book Arc】 - **Opening (~0%–15%)**: Frames the core problem — raw ingestion, transformation, and business-ready outputs must stay separated yet connected — and introduces the Four-Gate Governance Model (Boundary, RB-30, Evidence, AI proposal) as the safety net around medallion layers. - **Early (~15%–35%)**: Walks the architectural lineage from data warehousing through data lakes to lakehouse and data mesh, then maps the book's structure: ROI and integration strategy, multi-cloud medallion design, and scalable pipeline construction. - **Middle (~35%–55%)**: Deepens governance and compliance — data vault for auditable integration, evidence trails that answer auditors quickly, and the shift from "who wrote this change?" to "is this change safe?" as AI-generated changes become routine. - **Late (~55%–80%)**: Extends the same gates to MLOps (SageMaker workflows, model drift detection, rollback) and DevOps/CI-CD for data teams (GitHub Actions, automated testing, remote collaboration). - **Ending (~80%–100%)**: Covers phased cloud migration and coexistence, then platform optimization — performance tuning, cost control, observability, Delta Lake compaction/Z-ordering, and emerging trends. 【Key Takeaways】 - **Governance is the load-bearing wall, not overhead** (Early): The book's central argument is that medallion layers without governance are fragile — a bronze schema change can cascade to gold, and a failed transformation can corrupt months of curated data. - **Four gates, three outcomes** (Middle): Every change — human or AI-generated — passes Boundary, RB-30 (rollback within 30 minutes), Evidence, and AI proposal gates, each returning Go, Retry, or Kill. This is the book's most portable framework. - **The "who wrote this?" question is obsolete** (Middle): As AI assistants generate code and transformations, trust must attach to the change's safety properties (boundaries, reversibility, evidence) rather than the author's seniority. - **Audit readiness is an engineering requirement** (Middle): Consolidated evidence trails turn multi-day compliance excavations into minutes-long lookups — a design constraint, not an afterthought. - **MLOps inherits the same gates** (Late): Model deployment, drift detection, and rollback are treated as governed changes flowing through the same pipeline discipline as data transformations. - **Migration is a coexistence problem** (Ending): Phased and parallel approaches, downtime minimization, and risk mitigation matter more than cutover speed. - **Optimization is continuous** (Ending): Performance tuning, cost strategy, observability, and table-format maintenance (compaction, Z-ordering) keep the platform viable after launch. - **Fusion-Mart runs throughout** (Early–Ending): A fictional retailer's inventory, fraud, and migration scenarios anchor chapters 2–10, with hands-on labs producing evidence files that mirror production artifacts. 【Reading Tips】 - **Read the Reading Guide and Preface first** (around 30–35%): They explain the three questions and how chapters map to gates — this orientation makes the rest navigable. - **Deep-read Chapters 2, 5, and 6** if governance is your priority; they establish the boundary, RB-30, and evidence gates respectively. - **Skim the historical evolution chapter** if you already know the warehouse-to-lakehouse story; extract the governance-agility conflict argument and move on. - **Do the labs** (Chapters 2–10, GitHub Codespaces): They produce evidence artifacts, which is where the book's practical value concentrates. - **Treat MLOps and DevOps chapters as integration patterns**, not standalone tutorials — their value is showing how existing gates absorb new workload types. 【Coverage Limits】 The excerpts cover the book's framing, table of contents, governance model, and architectural history in detail, but do not include the actual technical content of the pipeline, MLOps, migration, or optimization chapters — specific tool configurations, code, and lab steps are not represented here.