Define the degradation ladder for an e-commerce site under extreme load. What is shed, in what order, and who decides?

What the interviewer is testing

Whether you treat shedding priority as a business decision made in advance rather than a technical default.

The ladder

Ordered by distance from revenue, shed from the top:

Recommendations and personalisation. Serve a static popular-items list. Nobody abandons a purchase because the recommendations are generic.

Reviews and ratings, or serve a cached aggregate rather than live content.

Non-essential enrichment — stock levels at nearby stores, delivery estimates for alternative options, related content.

Search refinements and faceting, falling back to simpler ranking.

Analytics and tracking events, buffered rather than sent synchronously. These must never be in the critical path in the first place.

Account features — order history, wishlists, saved items — degraded to read-only.

Never shed: browse the product, add to basket, checkout, payment. These are the revenue path.

The mechanics

Trigger on the right signal. Concurrency in flight and queue depth predict collapse; CPU utilisation lags it.

Reject early and cheaply, at the edge or on admission — a rejection costing as much as serving has not helped.

Shed requests whose deadline has already expired, which is free capacity nobody is waiting for.

Return a clear signal so well-behaved clients back off rather than retrying immediately.

Who decides

Commerce and product, not engineering. The ordering encodes what the business values, and engineering guessing at it produces a ladder nobody has agreed to. The conversation is short and it must happen before the event, not during.

What a strong answer adds

Test it. A shedding mechanism that has never fired under real conditions usually has a bug — an incorrect priority classification, a rejection path that is itself expensive, or a threshold that never triggers. A load test past capacity is the only way to know.

And instrument what was shed and why, so the business can see the cost of the capacity decision and decide whether to buy more headroom next year.

Common weak answers

Random shedding. Autoscaling as the answer, which takes minutes and may be quota-constrained exactly when needed.