Incident response, CI/CD design, Kubernetes and infrastructure trade-offs - practise the rounds an SRE or platform loop actually runs.
DevOps and SRE loops swap the algorithm round for a troubleshooting round, and it is usually the one that decides the outcome. You are handed a symptom - latency spiked at 3am, pods are restarting, the deploy went out and error rate climbed 20 minutes later - and scored on how you narrow it down. Jumping to a cause without checking a signal is the classic miss.
The design round is infrastructure-shaped rather than application-shaped. Expect to design a deployment pipeline with safe rollback, a multi-environment Kubernetes setup, a logging and metrics stack with a cost ceiling, or a disaster recovery plan with stated RPO and RTO targets.
Reliability questions are trade-off questions. There is no credit for "make it highly available"; there is credit for knowing what an extra nine costs in infrastructure and on-call load, and for saying which failure modes you deliberately chose not to protect against.
45-60 min
A live incident scenario. Narrow it down from symptoms using the signals you ask for.
60 min
Pipeline, cluster topology, networking, secrets and rollback strategy.
45 min
Automation scripting, log parsing, or a small tool. Usually lighter than a SWE DSA round.
45 min
A real incident you owned: detection, mitigation, and what the postmortem changed.
Narrow the blast radius before guessing at a cause. Establish what changed and when, whether it affects one service or many, one region or all, and which signal you would look at first. Interviewers are scoring your search strategy, not whether you name the right cause in the first minute.
Enough to automate. Expect scripting, log processing or a small CLI rather than hard algorithms - though Google SRE loops in particular still run a real DSA round at close to SWE difficulty.
Whether you can price reliability. Strong answers state an SLO, work out the error budget it implies, and say which failure modes are worth engineering against at that budget and which are not. Answers that treat availability as free read as inexperienced.
Run an infrastructure design mock and get scored on trade-offs, not buzzwords.