situational Medium GeneralSoftware Engineer

Halfway through a phased production rollout, your monitoring shows error rates climbing for the newly deployed service. What would you do?

How to handle this hypothetical in an interview: a worked example answer, the missteps that signal poor judgment, and the follow-up questions that test your reasoning.

A strong example answer

Halfway through a rollout of a payments service, our canary caught a spike in errors and I had it rolled back within minutes. Users never noticed. That outcome is the template: climbing error rates mid-rollout are an incident first and a release problem second. The moment monitoring moves, I'd check the blast radius — how much traffic sits on the new version, which endpoints fail, whether anyone actually feels it. If the impact is real, halt the rollout and roll back to the last stable version. No waiting for a perfect root cause. A rollback is cheap; letting a bad deploy bake while I debug is not.

Diagnosis comes after users are safe. In that incident the culprit was a missing database index that only mattered under production load. We fixed the migration, added a load test covering that query path, and wired automated canary thresholds so the next bad deploy rolls itself back. Throughout, I'd keep a short note running in the incident channel, then send a follow-up with cause and prevention once resolved. Users forgive a brief, well-communicated blip. They don't forgive finding out from their own errors.

A framework for your own answer

  • Situation: Frame the rollout stage, the metric that alarmed you, and who is affected.
  • Task: State your two goals: protect users first, then diagnose.
  • Action: Walk through rollback, root-cause analysis, and communication in order.
  • Result: Quantify the impact window and the prevention you added afterward.

Common mistakes to avoid

  • Don't debug in production while users feel the errors — roll back first, then investigate on a stable system.
  • Hoping the errors settle on their own is not a plan; the dashboard is already telling you something is wrong.
  • Silence is its own failure, because support and stakeholders should never learn about an incident from angry users.
  • What stops the next bad deploy? An answer that ends at the bug fix, with no test, alert, or automation, leaves the door wide open.

Follow-up questions to prepare for

Expect the interviewer to dig past your first answer. Common follow-ups for this question include:

  • What rollback strategy do you prefer and why?
  • How would you handle this if the deploy could not be rolled back cleanly?
  • Who do you notify during an incident, and in what order?

Related practice questions

Keep drilling with these, or browse the complete question bank.

Next up

Practice this question on the clock

Reading an answer and delivering one are different skills. Run a timed round and rate yourself.

Start a practice round

Next up

Interview coming up?

Have your post-interview follow-up email ready before you need it.

Follow-up email generator