A stakeholder requires 99.99% availability. Your service depends on five internal services each offering 99.9%. What do you say?

What the interviewer is testing

Whether you do the arithmetic and turn an impossible commitment into a costed engineering conversation.

The arithmetic

Serial hard dependencies multiply: 0.999^5 ≈ 99.5%, before your own service's code, infrastructure or deployments contribute anything.

That is roughly 43 hours of unavailability a year against a target of 52 minutes. The target is unreachable as designed, by a wide margin.

The options, with what each costs

Reduce hard dependencies in the request path. Almost always the highest-leverage move, and usually several of the five are not genuinely required to serve a request.

Convert hard dependencies to soft ones. A hard dependency's failure fails the request; a soft one degrades it. Cache the reference data, default the enrichment, queue the non-urgent write. Each conversion removes a term from the multiplication.

Add redundancy where failures are genuinely independent — which for shared internal services they usually are not, so this is often less effective than it appears.

Raise the dependencies' targets, which means investment in five other teams and needs justification across all their consumers.

Lower the target to what the architecture supports.

The conversation to have

Ask what the target is actually for. 99.99% is frequently chosen because it sounds appropriately serious rather than derived from a business need. The questions that produce a real number: what does an hour of downtime cost, what do competitors offer, and does this apply to every journey or only the critical ones?

Availability targets almost always differ by journey, and a single system-wide number is usually wrong for everything.

What a strong answer adds

Presenting it as a menu with costs: "99.5% is what we have; 99.9% requires making these three dependencies soft, about six weeks; 99.99% requires multi-region active-active and this ongoing cost. Which do you want to buy?" That converts an argument into a decision.

Common weak answers

Accepting the target and hoping. Refusing without offering the path to each level.