Design Lab — Match the data store to the thing you cannot change later
AZ-305 › Unit 2 › Design data storage solutions for relational data
Design Lab — Match the data store to the thing you cannot change later
Lab brief
- 2 — Design data storage solutions
- 20–25%
- 20 minutes
- $0 — no subscription needed
- Core
Data questions punish the same instinct twice: picking the most modern option, and picking one you can change your mind about later. Some choices here are cheap to revise — a tier, a throughput setting — and some are effectively permanent, like a partition key or a migration that had to rewrite application code. The exam tests whether you can tell which is which.
Read the brief, commit before you reveal, then check.
The brief
Woodgrove Logistics — data platform selection
An on-premises SQL Server estate: twelve databases that query each other, forty SQL Server Agent jobs driving nightly settlement, and two linked servers to a partner system. The team wants PaaS and does not want to rewrite the application.
Commit before you reveal
▶The one-way doors in this lab — open after committing
Two of the four decisions below are reversible with a slider. Two are not: a database platform that forced an application rewrite, and a partition key, which is immutable and requires data migration to change. Knowing which decisions are one-way is most of what "design" means on this exam.
Question 1 — the estate that must not be rewritten
Twelve databases that query each other, forty SQL Server Agent jobs, and two linked servers — moving to PaaS without an application rewrite. What should the design name?
Question 2 — the choice you cannot take back
The engineer proposes parcelStatus (six values) as the partition key. Which statements are true and decide whether that is acceptable? Select all that apply.
Select all that apply
Question 3 — freshness versus staying up
Tracking reads may be a second or two stale; write latency and surviving a regional problem matter more. Which consistency choice fits, and why?
Question 4 — the order that finds the one-way doors first
Order the data store selection method so that irreversible choices surface before reversible ones.
Put the steps in order
- 1Choose the platform that supports the required surface without a rewrite
- 2Fix the keys and partitioning, and flag anything immutable in writing
- 3Read the ACCESS pattern: what is read, how often, how fresh it must be
- 4Tune the reversible knobs — tier, throughput, replicas
Why the runner-up loses
Four homes for data, and the requirement that picks each
| Attribute | Recommended | |||
|---|---|---|---|---|
| Buy it when | A new app, one database at a time | An estate that must move unchanged | You need the OS or an unsupported feature | Global writes and flexible schema |
| Instance-scoped surface | No — database-scoped | Agent, cross-database queries, linked servers | Everything SQL Server has | Not applicable |
| Who patches it | Azure | Azure | You | Azure |
| The irreversible choice | Rewrites forced by missing surface | Instance size and subnet | The operating burden you accepted | The partition key |
| Reversible later | Tier, pool, replicas | vCores, storage | VM size, disks | Throughput, consistency, regions |
The row that decides the migration question is the second: database-scoped versus instance-scoped. Anything the application does across databases — a job, a cross-database join, a linked server — lives on the instance, and no amount of elastic pooling puts it back.
The shape of the answer
The traps that catch people
The method, for any question of this shape
How to attack a data question in 90 seconds
Separate the migration from the new build
A scenario with an existing estate and a new workload is two questions. Answer them independently; the right answer is often one PaaS relational platform AND one non-relational store.
Before you move on
Return cold after the Unit 2 lessons. If you can name, for each decision in this lab, whether it is a slider or a one-way door, you have the habit this whole unit is trying to build.