Interview questions for a site reliability engineer.

The questions.

1. 01

   technical medium · cicd

   Design a deploy pipeline for a service that handles payments.

   Strong answer

   Build → test (unit, integration, contract) → staging → canary with metrics-gated rollout → full deploy. Knows when to use feature flags vs deploys. Rollback is automatic.

   Weak answer

   Push-to-main-deploys-to-prod. No canary. No rollback strategy.
2. 02

   technical medium · iac

   Terraform vs Pulumi vs CDK — how would you pick?

   Strong answer

   Team language fluency, existing ecosystem, need for programmability, drift-detection. Not dogmatic.

   Weak answer

   'Terraform because it's standard.' Or 'Pulumi because code is better than HCL.'
3. 03

   communication medium · exec

   Explain the cost of an on-call rotation to a non-technical CFO.

   Strong answer

   Concrete dollar math — engineer hourly × expected rotations × burnout attrition cost. Knows the true cost isn't 'the pager.'

   Weak answer

   'On-call costs money.' No concrete framing.
4. 04

   situational medium · oncall-culture

   Your team's on-call is burning people out. You own it. What do you change in the next month?

   Strong answer

   Reduces noise (alert hygiene). Measures toil. Redistributes load. Caps consecutive rotations. Doesn't just add more people.

   Weak answer

   'Hire more engineers.' Or 'make people tougher.'
5. 05

   technical medium · observability

   What's the first dashboard you'd build for a new service?

   Strong answer

   Golden signals — latency, traffic, errors, saturation. Per-endpoint where relevant. SLO framing.

   Weak answer

   CPU and memory. No user-facing metrics.
6. 06

   culture medium · blame

   A junior engineer ran a query that took down prod for 15 minutes. How do you respond?

   Strong answer

   Blameless. Focuses on why it was possible (no safeguards, bad tooling). Fixes the system. Reassures the engineer. Doesn't shield them from the learning.

   Weak answer

   Either fires them or coddles them. Either way, system unchanged.
7. 07

   technical hard · incident

   Describe the worst production incident you've been on-call for. Timeline, what you did, what you changed afterward.

   Strong answer

   Honest retrospective. Technical + human factors. Clear systemic change — runbook, process, or code. Names people specifically.

   Weak answer

   Vague. Blames 'the old team.' No systemic change.
8. 08

   technical hard · kubernetes

   You have a K8s pod that's being OOMKilled intermittently. How do you diagnose without just bumping memory limits?

   Strong answer

   Checks actual memory usage pattern (profiler, metrics). Finds the root cause — leak, spike, burst. Considers JVM/runtime overhead. Sets requests/limits intentionally.

   Weak answer

   Bumps memory. Or 'add more replicas.'
9. 09

   experience baseline · resume-probe

   Your resume says you migrated from EC2 to Kubernetes. Team size, timeline, what broke?

   Strong answer

   Specifics. Something specific that broke (networking, secrets, state). Honest about why it took longer than planned.

   Weak answer

   Paraphrases the resume. Nothing broke.
10. 10

    technical medium · security

    Walk me through how you'd set up secrets management for a team of 30 engineers.

    Strong answer

    No secrets in repo ever. Vault / AWS Secrets Manager / SOPS. Dev gets different secrets than prod. Rotation is automated. Audit trail.

    Weak answer

    '.env files we share on Slack.' Or 'engineers memorize prod creds.'