How to answer: “How would you debug a production issue?”

What they’re actually asking

They're simulating the worst hour of the job and watching for panic. The wrong answer dives into code. The right answer establishes impact, stops the bleeding, then hunts — in that order. Method under pressure is the whole test.

How to structure your answer

Give your sequence: confirm scope and impact first, check what changed last (deploys, configs, dependencies), mitigate before root-causing — roll back, feature-flag off, scale up — then use logs and metrics to isolate. End with the unsexy part: the postmortem and the fix that makes it impossible to repeat.

Example answer

“First question is who's affected and how badly — one customer or all of them changes everything. Then what shipped recently; most incidents are the last deploy. If rollback is safe, I roll back before I understand the bug, because users don't care about root cause. Once traffic is healthy I trace it properly. Last one I handled was a checkout outage: rolled back in 8 minutes, found a null in a currency field that night, and added the validation that makes it unrepeatable.”

What sinks people

• Jumping straight to reading code while production burns
• No mention of communicating status to anyone during the incident
• Ending at "found the bug" — no rollback plan, no postmortem, no prevention