Design and Implement Data Storage
Core concepts
Azure storage for analytics centers on Azure Data Lake Storage Gen2 (ADLS Gen2), which layers a hierarchical namespace over Blob Storage. Understand the difference between flat blob namespaces and true directories: ADLS Gen2 enables atomic directory renames and efficient partition elimination. Know how to choose between Parquet, Delta Lake, Avro, and ORC formats, and when to use partitioning versus bucketing. PolyBase and Synapse external tables require specific file layouts and statistics. Also cover Azure Synapse Analytics table distributions (hash, round-robin, replicated) and indexing (clustered columnstore, heap). For streaming, understand Event Hubs Capture to ADLS Gen2 and Azure Stream Analytics output partitioning.
Traps
Questions often test whether you know that hash-distributed tables improve joins on large fact tables but hurt small dimension tables; replicated tables are best for small dimensions. Another trap: choosing rowstore vs columnstore for small tables or staging. Also watch for partition switching requirements (same filegroup, aligned partitions) and external table limitations (no statistics auto-create).
Worked example
A fact table of 5 TB joins to a 50 MB dimension. Rule of thumb: hash-distribute the fact table on the join key, replicate the dimension. That avoids shuffling the large table and keeps the small table local to each compute node.
Key rule
Always match the storage layout to the query pattern: partition by date for time-filtered queries, use columnstore for large scans, and avoid over-partitioning (thousands of small files hurt performance).