Design an ML Job Scheduler
Two Netflix candidates reported this prompt in 2026. One sat it in June as the system design half of the first virtual onsite. The other got it in September, in the second round of a loop for what they believed was a data platform team. Neither report names a level, and neither gives exact wording, so treat the prompt above as the shape of the question. The June report is unusually clear about how the hour went. The interviewer worked down a list of topics in order. It covered the kinds of ML jobs, what each kind needs from the scheduler, the allocation policy, how the scheduler watches the cluster, crash handling for every component, and a short stop on multi-tenancy. He went deep on each one and interrupted often. Once he had what he needed, he moved on.
Most of the hour goes on policy. Even with generous assumptions the traffic is tiny, on the order of ten thousand job submissions a day. What makes it hard is that the jobs want very different things from the same hardware. That hardware is expensive enough that idle time is real money, and GPUs fail often enough that a long training run will almost certainly see one die.
NetflixPremium lesson · System design round
The Netflix design, worked end to end.
- Requirements and scopeSystem design · Frequently asked
- The API and data model
- The design, drawn out1 diagram
- 5 deep divesWith the tradeoffs
$49 a year · all 107 companies · instant access
181 Netflix interview reports · guide updated
Netflix interview guideKnow what Netflix asks before you walk in.
20 coding · 9 system design · 19 behavioral questions
Get their questions first.
InterviewLegend is independent and not affiliated with or endorsed by any company it covers.