MidTroubleshooting

Your load balancer shows 5xx errors jumping from 0.1 percent to 12 percent of requests over ten minutes. There was no deploy. What do you do?

What they are really testing: Whether you can triage without the comfort of a recent deploy to blame, and whether you know the difference between the load balancer failing and the target failing.

A real interview question

Your load balancer shows 5xx errors jumping from 0.1 percent to 12 percent of requests over ten minutes. There was no deploy. What do you do?

What most people say

drag me

I would restart the service, that usually clears 5xx errors, and then see if it comes back.

Restarting is a coin flip that destroys the evidence you needed, it shows no model of where the error is generated, and it gives you nothing to say when the errors return twenty minutes later.

The follow-ups they ask next

  • The 5xx are all 502s. What does that tell you?

    The target closed the connection or returned a malformed response. Look for OOM kills, keepalive mismatch between the ALB idle timeout and the app timeout, or crashing workers.

  • How do you decide whether to page the downstream team?

    Page them once you have one piece of evidence, not a suspicion. Bring the latency graph and a request id, and say what you have already done to contain it.

What the interviewer is listening for

  • Distinguishes load balancer 5xx from target 5xx
  • Adds capacity as a containment move before root causing
  • Considers non deploy changes like secret rotation or certificate expiry
  • Contains a failing dependency instead of trying to fix someone else's service at 2am

What sinks the answer

  • Restarts everything as the first and only move
  • Treats all 5xx as one undifferentiated blob
  • Never checks whether errors are concentrated on one host or one route
  • Declares victory the moment the graph dips for 30 seconds

If you genuinely do not know

Say this instead of freezing. Reasoning out loud from what you do know beats silence every single time, and a good interviewer is listening for exactly that.

I have not owned an ALB in production, but the reasoning I would use is: add capacity to contain it, then work out whether the error is being generated by the load balancer or by my own app, because those lead to different places.

Keep going with troubleshooting

All 336 cloud engineer questions

Knowing the answer is not the same as recalling it under pressure

Sign in to send the questions you fumble to spaced recall, so they come back right before you would forget them, and learn the concepts behind them with hands-on labs.

Start free