Storage choices: free-recall review
Free-recall prompts across the storage unit. Answer each in your own words first — the SQL/NoSQL trade, access-pattern design, the metadata/bytes split, durability vs availability, time-series and search internals, and the dual-write fix — then reveal and compare.
Retrieval beats re-reading. For each prompt, say or write a full answer from memory before you open the model answer. The effort of reconstructing the SQL/NoSQL trade, the metadata/bytes split, and the dual-write fix is what makes them stick.
Reconstruct the unit’s spine without looking back: when relational vs each NoSQL family wins, how access-pattern-first design cuts both ways, why blobs never live in the database, how durability differs from availability, what makes time-series and search stores specialized, and how to keep two stores in sync.
- 01Name the four NoSQL families and the access shape each is built for, and the ACID/BASE trade.
- 02Why does access-pattern-first design cut both ways, and what is polyglot persistence?
- 03Why never store blobs in the DB, and state the metadata/bytes split with presigned URLs.
- 04Distinguish durability from availability, and explain lifecycle tiering.
- 05What makes time-series and search stores specialized, and how do you fix the dual-write problem?
If you could reconstruct each answer from memory, you hold the unit’s spine. Relational (ACID) is the default for invariants the engine must enforce; the four NoSQL families (document, key-value, wide-column, graph) scale out and are fast for a known access pattern, trading ad-hoc flexibility and, under BASE/CAP, strong consistency. Access-pattern-first design is powerful when patterns are known and brutal when they change, which is why mature systems go polyglot. Large bytes never live in the database — metadata in the DB, bytes in object storage, served by presigned URLs so the app tier is never a pipe — and durability (bytes exist, ~11 nines) is not availability (reachable now), with lifecycle tiering tracking cost to coldness. Time-series stores (LSM, downsampling, retention) and search engines (inverted index) are specialized for write shape and text, added on a measured limit and treated as derived indexes whose dual-write consistency is solved by deriving from one authoritative commit (CDC or outbox), never two independent writes. The thread: storage design is choosing which guarantees you need and which costs you can pay.
Something unclear?
Ask a question about this lesson. Questions are anonymous and go straight to the author to make the lesson better.