Memory hook: Choose object, file, key-value, document or graph by the required access pattern.
Must remember
- Blob Storage stores objects in containers and supports tiers with different cost/access behaviour. Data Lake Storage adds hierarchical namespace capabilities for analytical workloads. Azure Files exposes shared file access; Table Storage provides a key/attribute-style non-relational store.
- Cosmos DB supports distributed database workloads through supported APIs/models. Partition keys determine distribution and query locality; a poor key can create hot partitions or expensive cross-partition work. Request Units measure resource consumption of operations.
- Consistency choices trade read guarantees, latency and availability under the selected deployment. Strong, bounded staleness, session, consistent prefix and eventual describe different guarantees where supported. Session consistency preserves relevant session-level behaviour; it is not globally strong consistency.
- Documents can vary in fields, but applications still need validation and evolution rules. Graphs emphasise relationships/traversal; key-value stores emphasise direct key access. Flexible schema does not mean no data modelling.
- Replication, backup, encryption and access are separate decisions. Global distribution must match residency rules and write/conflict requirements. A non-relational system may support transactions within particular boundaries; never assume every transaction spans all partitions globally.
Choose under exam pressure
| Requirement | Choice and reason |
|---|---|
| Store videos with metadata and lifecycle rules | Blob storage plus suitable metadata/indexing. |
| Application needs key/document queries across Regions | Evaluate Cosmos DB with proper partitioning. |
| Existing app expects an SMB share | Azure Files where compatible. |
Traps
- NoSQL does not mean no schema discipline.
- Global replication does not remove latency/consistency trade-offs.
- Request Units are not simply document count.
Active recall
1. Why does partition-key choice matter?
It affects load distribution, query locality and scaling.
2. Which storage interface best fits unmodified file-share access?
A supported file service such as Azure Files.
3. Is session consistency identical to strong consistency?
No. Its guarantee is scoped to session behaviour rather than all reads globally.
4. Why can two same-sized documents cost different RUs to query?
Indexing, query complexity, consistency and access pattern affect resource use.
5. What should be checked before multi-Region writes?
Conflict handling, consistency, latency, residency and supported service configuration.