Your platform abstracts Kubernetes entirely — teams write a short manifest and everything is generated. Now every production incident escalates to the platform team. What is happening?
Show the full answer Hide the answer
What the interviewer is testing
Whether you understand that abstraction height is a staffing decision as much as a design one.
What is happening
Abstractions leak under failure. During normal operation the abstraction holds and teams are productive. When something breaks, the diagnosis requires understanding the layer beneath — and the teams have deliberately been prevented from learning it.
So the platform team becomes the escalation path for every team it abstracted for, and that load scales with the number of consuming teams while the platform team does not.
This was predictable and is usually absent from the business case, which counts the productivity gain and not the support obligation.
The options
Invest in failure-mode transparency rather than lowering the abstraction. The goal is that teams can diagnose within the abstraction's own vocabulary: surface the generated resources read-only, translate underlying errors into terms the manifest uses, and provide a debug view showing what the platform did with the manifest and why. Most escalations are "I do not know what happened", not "I cannot fix it".
Provide a supported escape hatch, not a cliff. Teams that need to see the layer beneath should be able to, with the platform's blessing, for diagnosis at minimum.
Staff for it deliberately. If the abstraction is genuinely worth keeping, the support load is part of its cost and needs a rota, an SLO and headcount — not the platform engineers' evenings.
Teach. Some teams should learn the underlying platform, particularly those on call for high-consequence services.
The judgement
A high-level default covering most services, with a documented and supported route to the layer beneath, is where successful platforms land. The failure is treating the abstraction as a wall rather than as a default.
What a strong answer adds
Measuring escalation volume and cause as a platform product metric — it points precisely at where the abstraction leaks, and those are the highest-value transparency features to build.
Common weak answers
Removing the abstraction, discarding real value. Writing more documentation, which does not help someone debugging at 3 AM.