Zoom DevOps / SRE Interview Questions

1.Tell me about a specific incident where you drove a measurable improvement in service reliability. What was broken, what did you change, and how did you know it was better afterward?

2.Walk me through a time you improved system throughput or reduced latency specifically in response to a traffic spike or capacity event — not planned optimization, but an unplanned scaling situation.

3.Walk me through a time you reduced latency or improved throughput on a high-traffic service. What were the numbers before and after, and what tradeoffs did you accept to get there?

4.Tell me about a time you used ML or AI tooling — whether a model you deployed, an AI-assisted workflow you adopted, or an AI feature you helped ship — and what actual impact it had.

5.Describe a production concurrency bug you debugged — something like a race condition, deadlock, or thundering herd. How did you reproduce it, and what did the fix look like?

6.Tell me about a time you pushed back on shipping a change because you thought it was a reliability risk — and the team disagreed with you. How did it play out?

7.Walk me through how you'd set up a Kubernetes liveness and readiness probe for a stateful media relay service. What's the difference, and why does it matter?

8.You need to add a new environment variable to 200 microservices deployed across our Kubernetes clusters. What's your approach, and what could go wrong?

9.Our Kafka consumer lag on the meeting-events topic has been growing for the past 2 hours and is now at 10 million messages. How do you triage and recover?

10.We want to run chaos experiments against our meeting signaling stack. How would you design the first three experiments, and what guardrails would you put in place before running them in production?

11.Design a deployment pipeline for Zoom's native desktop client — Windows and macOS — that needs to ship a hotfix to 100 million endpoints within 24 hours without breaking active meetings. What does that pipeline look like?

12.Our global media edge network needs to route a new meeting participant to the nearest media node. The routing service is seeing elevated P99 latency — 180ms — on the node-selection call, which is supposed to complete in under 20ms. How do you diagnose and fix this?

13.Design a health-dashboard service that gives Zoom's on-call engineers a real-time view of meeting reliability across all regions — things like join success rate, audio/video failure rates, and media node availability. What does the data pipeline and UI look like?

14.Zoom Phone runs as a cloud PBX where call-state — active calls, hold, transfer — has to survive the failure of the service instance handling it. How would you design a call-state store that's both highly available and consistent enough that a user's active call never silently drops mid-transfer?

15.We want to give Zoom AI Companion the ability to join any in-progress meeting and start generating a live transcript within 10 seconds of being invoked. Design the infrastructure that provisions and connects that transcription workload on demand.

16.Design a multi-region config-delivery service that pushes feature flags and runtime configuration to every Zoom microservice — roughly 500 services, thousands of pod instances — within 30 seconds of a change, while staying available even during a partial AWS region failure.

17.Design the on-call incident management pipeline for Zoom Meetings — from the moment an alert fires to the moment a post-mortem is published. Assume incidents can range from a single-region media outage to a global control-plane failure affecting millions of concurrent meetings.

18.Zoom's media edge nodes need to be updated with a new kernel security patch, but they're actively handling live meeting traffic 24/7. Design a zero-downtime patching pipeline for roughly 2,000 edge nodes distributed globally, including how you handle a patch that causes packet loss or elevated jitter on the first nodes it touches.

19.Tell me about a time you had to earn the trust of an on-call rotation you were new to. What did you do in the first few weeks, and how did the team respond?

20.Describe a time you caught a gap in your team's observability — something important that nobody was alerting on. How did you decide it was worth fixing, and what did you actually build?

21.Tell me about a time you had to push a reliability improvement that required getting engineers outside your team to change how they deployed their services. How did you get traction?

22.You're mid-incident on a Zoom Meetings outage and your runbook is wrong — the documented recovery step makes things worse. How do you lead the response from that moment forward?

23.Tell me about a time you identified that your team's deployment process was creating toil — repetitive, manual, error-prone work — and led the effort to eliminate it. What was the before state, what did you build, and what resistance did you encounter?

24.Describe a time you disagreed with a product or engineering team about whether a service was ready to handle a major traffic event — a product launch, a large conference, or a viral moment. How did you make the case, and what happened?

25.Estimate the total storage cost Zoom incurs per day just from Zoom AI Companion meeting summaries. Walk me through your assumptions.

26.Zoom Meetings join latency — the time from clicking 'Join' to seeing participants — spiked 40% last Tuesday from 2 PM to 4 PM UTC. No deploys happened. How do you structure your investigation?

27.We want to sunset a legacy internal deployment tool that roughly 80 Zoom engineering teams still use for their release pipelines. How would you plan and execute that migration without creating a wave of outages?

28.Estimate how many Kubernetes pods Zoom needs cluster-wide to serve Zoom Phone at peak load. Walk me through your reasoning.

29.Design a cost-attribution system that tells each of Zoom's 500+ microservice teams exactly how much their service is spending on AWS each month — broken down by compute, data transfer, and storage — with enough accuracy that teams will act on it.

30.Zoom's SLO for meeting join success rate is 99.9%. Estimate what that SLO costs the business in failed joins per day, and then tell me how you'd determine whether 99.9% is the right target or whether we should invest to reach 99.95%.