LFC Architecture 2 — Questions and Answers
Question 1: In a Lakehouse architecture, what role does the metadata layer play?
- It stores raw binary data for fast retrieval
- It provides schema, versioning, and ACID transaction management over object storage (Correct answer)
- It compresses Parquet files before writing to cloud storage
- It routes SQL queries to the appropriate compute cluster
Correct answer: It provides schema, versioning, and ACID transaction management over object storage
The metadata layer (e.g., Delta Lake, Iceberg, Hudi) tracks schema evolution, file statistics, and transaction logs to enable ACID semantics on object storage.
Question 2: Which characteristic distinguishes a Lakehouse from a traditional two-tier Lambda architecture?
- The Lakehouse uses only streaming ingestion while Lambda uses only batch
- The Lakehouse eliminates the separate serving layer by enabling direct BI access on the storage tier (Correct answer)
- The Lakehouse requires a dedicated OLTP database for transactional workloads
- The Lakehouse stores data exclusively in columnar in-memory formats
Correct answer: The Lakehouse eliminates the separate serving layer by enabling direct BI access on the storage tier
A Lakehouse collapses the separate data warehouse serving layer so BI tools can query the same open-format storage used for ML and data science.
Question 3: What is the purpose of Z-ordering (data skipping) in Delta Lake's storage layer?
- To encrypt data files at rest using AES-256
- To co-locate related data within Parquet files so queries scan fewer files (Correct answer)
- To replicate data across multiple cloud regions automatically
- To convert JSON ingestion files into columnar Parquet format
Correct answer: To co-locate related data within Parquet files so queries scan fewer files
Z-ordering multidimensionally sorts data so that frequently filtered columns are co-located, enabling Delta Lake to skip irrelevant files during query execution.
Question 4: In Databricks Lakehouse architecture, what does the concept of 'medallion architecture' describe?
- A GPU-accelerated compute tier for deep learning workloads
- A multi-hop data refinement pattern using Bronze, Silver, and Gold layers (Correct answer)
- A disaster recovery strategy that replicates Delta tables across availability zones
- A billing model that awards compute credits based on query performance
Correct answer: A multi-hop data refinement pattern using Bronze, Silver, and Gold layers
The medallion architecture organizes data into Bronze (raw), Silver (cleaned/conformed), and Gold (business-aggregated) layers for progressive data quality.
Question 5: Which component in the Databricks Lakehouse platform serves as the unified governance layer for data and AI assets?
- MLflow Model Registry
- Unity Catalog (Correct answer)
- Delta Live Tables
- Databricks SQL Warehouse
Correct answer: Unity Catalog
Unity Catalog provides centralized access control, auditing, lineage, and data discovery across all workspaces in a Databricks account.
Question 6: What does 'schema enforcement' in Delta Lake prevent?
- Queries from reading encrypted columns without proper IAM roles
- Writing data whose schema does not match the table's registered schema (Correct answer)
- Auto-scaling clusters from spinning up additional worker nodes
- Time travel queries from accessing snapshots older than 30 days
Correct answer: Writing data whose schema does not match the table's registered schema
Schema enforcement rejects writes that contain columns not present in the target Delta table's schema, protecting data integrity.
Question 7: In a Lakehouse, compute and storage are typically described as being decoupled. What is the primary operational benefit of this design?
- Data is automatically deduplicated across all storage buckets
- Compute resources can be scaled independently and shut down without losing data (Correct answer)
- SQL queries are automatically converted to Spark jobs for parallelism
- Object storage buckets are billed at the same rate as SSD-attached volumes
Correct answer: Compute resources can be scaled independently and shut down without losing data
Decoupling compute from storage means you pay only for compute when jobs run, and idle clusters can be terminated while data persists indefinitely in object storage.
LFC Architecture 2
In a Lakehouse architecture, what role does the metadata layer play?