KPMG Data Engineer Interview Questions (2026)

Data Warehouses are structured, schema-on-write systems optimized for high-performance SQL analytics, while Data Lakes store raw, diverse data with a flexible schema-on-read approach. Delta Lake is an open-source storage layer that enhances a data lake with data warehouse-like reliability, performance, and governance features. ### Mechanics and "Why" * **Data Warehouse**: These systems (e.g., Snowflake, Google BigQuery) enforce a schema *before* data ingestion (schema-on-write). Data is typically structured, cleaned, and transformed into a columnar format, optimized for analytical queries. They often employ Massively Parallel Processing (MPP) architectures and features like Snowflake's micro-partitions and clustering to ensure fast query performance for BI and reporting. Compute and storage are often tightly coupled or managed as a unified service. * **Data Lake**: A data lake stores vast amounts of raw, semi-structured, or unstructured data in its native format, typically on inexpensive object storage like AWS S3 or Azure Data Lake Storage (ADLS). It uses a schema-on-read approach, meaning the schema is applied at query time, offering immense flexibility for diverse data types (logs, IoT, multimedia). While cost-effective for storage, raw data lakes often lack built-in mechanisms for data quality, governance, and transactional consistency. * **Delta Lake**: Built on top of a data lake, Delta Lake adds a transactional storage layer that brings ACID properties (Atomicity, Consistency, Isolation, Durability) to data stored in formats like Parquet. It achieves this through a transaction log that records every change, enabling features like schema enforcement and evolution, time travel (accessing historical versions), and robust upserts/deletes. Delta Lake bridges the gap by offering data reliability and performance optimizations (e.g., Z-ordering for data skipping) within the economic flexibility of a data lake. ### Key Trade-offs The choice depends on data maturity, use cases, and budget. Data Warehouses excel for structured, curated data requiring high-concurrency SQL queries and predictable performance. Data Lakes are ideal for storing all data, regardless of structure, for exploration and machine learning. Delta Lake combines the best of both: lake economics for storage with warehouse-like reliability and performance for data engineering workloads. For instance, Delta Lake's `MERGE INTO` statement provides robust upsert capabilities, which are crucial for slowly changing dimensions and data synchronization, a feature often more complex in raw data lakes: ```sql MERGE INTO target_table AS t USING source_table AS s ON t.id = s.id WHEN MATCHED THEN UPDATE SET t.value = s.value WHEN NOT MATCHED THEN INSERT (id, value) VALUES (s.id, s.value); ``` While Data Warehouses offer built-in auto-tuning, Delta Lake relies on engines like Apache Spark or Databricks for processing, requiring more operational knowledge but offering greater control over complex transformations. In the interview, also mention the "Lakehouse" architecture, which Delta Lake enables, aiming to unify data warehousing and data lake capabilities into a single platform.