Site reliability engineer leadership interview questions

As asked

Your on-call rota is getting paged 40 times a week and people are burning out. How do you bring the load down without missing real incidents?

Sample answer outline

Treat alert hygiene as a programme, not a one-off. Categorise the last 4 weeks of pages: actionable vs auto-resolving, real impact vs symptom of a deeper issue, owned vs not owned. Delete alerts that auto-resolve and have no follow-up action. Merge duplicate alerts. Move symptom-based alerts to SLO burn-rate alerts. Push noisy alerts back to the team that owns the underlying system. Set a target alert budget per week and hold teams to it. Track it in retros.

Expect these follow-ups

• How do you stop a team from re-adding alerts you just removed?
• Which team owns an alert that fires from a shared system?
• What is the failure mode of burn-rate alerts?

As asked

You want to start a chaos engineering programme at a company that has never done one. What do you do in the first 6 months?

Sample answer outline

Build trust before doing anything destructive. Start with read-only game days: walk a team through a simulated incident on a whiteboard, see what falls out, fix the easy gaps. Move to staging-only fault injection: kill a pod, drop a percentage of network packets, simulate a slow database. Document hypotheses up front so the exercise can be falsified. Only after the staging muscle exists, run a controlled production experiment with a small blast radius and a hard abort switch. Pair with an SRE every time so the host team is not on the hook alone.

Expect these follow-ups

• How do you handle a VP who says chaos engineering is too risky?
• What is your blast radius policy for production experiments?
• What metric tells you the programme is working?