Skip to content
Interview rounds

Observability · Design under changing requirements

Visibility without touching the code

How would you use eBPF for visibility without instrumenting code, and what are its limits?

Platform-SREStaff10 min
Interview-style scenarioLast reviewed: Checked against official documentation

Tuesday 15:00 · platform planningIllustrative scenario, not a real incident

A slow checkout took four hours to trace to one service, because half the fleet reports nothing

The company runs forty services on Linux nodes in Kubernetes, written in five languages. About half carry a tracing SDK. The rest include a vendor binary and two services whose owners have left.

After last week's incident, leadership wants to see how many requests each service takes, how many fail and how long they take, plus a map of who calls whom. They want it within weeks, without asking forty teams to change code.

What the dashboards and logs show

Services
40, in Go, Java, Python, Node.js and C++
With a tracing SDK
19 of 40
Cannot be rebuilt
3 (one vendor binary, two without owners)
Deadline
weeks, not quarters

Clarify · optional · scored under "Hypothesis before change"

Ask the interviewer

0 of 3 asked

Each question reveals a fact. What you ask before you act is part of your score.

First move · scored under "Mitigate before root cause"

Another slow checkout could come this week, and half the fleet is still dark. What do you do first?

Your rubric score, the model answer, the follow-up questions and the related lessons appear once you have answered the last step.