A news organisation's site is functionally simple - editors publish articles and readers read them - yet its architecture is unusually elaborate. Which non-functional characteristic explains that, and what structure does it force?
Show the full answer Hide the answer
What is being tested
That you can look past the functional description and identify the quality attribute doing the structural work.
The reasoning
A newsroom might publish a few hundred articles a day. Those same articles might be read tens or hundreds of millions of times, and on an election night or during a major breaking story the read volume arrives as a step, not a ramp — there is no gradual warm-up in which to autoscale.
Two properties follow directly:
Reads and writes have almost nothing in common. The write path values editorial workflow, collaborative editing, embargoes, corrections, versioning and correctness. Volume is trivial. The read path values throughput and latency, needs almost no logic, and is dominated by the same small set of extremely popular objects.
Therefore they should not share a failure domain. If the publishing CMS is down, readers should still be able to read. Rendering articles to cacheable artefacts and serving them from a CDN gives you exactly that: the read path survives an origin outage, which is not a feature anyone requested but is the most valuable property the architecture has.
Where the interesting nuance sits
The hard part is not the static content — it is the parts that resist caching:
| Element | Why it resists | Typical answer |
|---|---|---|
| Personalised recommendations | Unique per user | Client-side call to a separate service; degrade to editorial defaults |
| Paywall and entitlement | Per-user state | Edge-evaluated token; fail open or closed by business rule |
| Live results and vote counts | Changes constantly | Short TTL plus stale-while-revalidate; a separate high-write path |
| Comments | Write-heavy and unpredictable | Fully separate service; the article renders without it |
Each of those is a decision about what to sacrifice when it fails, and each is answered by business requirement rather than by engineering preference. Serving a stale vote count for 15 seconds is fine. Serving the wrong paywall decision is not.
The transferable rule
When the read-to-write ratio is extreme and the read set is concentrated, precompute and push to the edge. When it is not, this architecture is expensive machinery for no benefit.