OpenAIVerified interview challenge
Sora 3: Design preemption and recovery for Sora jobs
Design how long-running GPU jobs can be interrupted and resumed when priorities change.
Unlock with Full AccessThe challenge
A video-generation cluster must make room for urgent work while other jobs are already running. Some jobs use several GPUs, and checkpoints are only possible at certain steps. Design how the system decides what to interrupt, preserves progress and resumes work without producing duplicate outcomes.
- Build on the existing job service
- Basic submission and queueing are already in place. Focus on interruption, recovery and execution ownership.
What you'll practise
Preemption decisions
Weigh the capacity released against the work lost when a job is interrupted.
Checkpoint and resume
Account for when progress can be saved and what happens if recovery fails.
Execution ownership
Keep results, charges and job status consistent when several attempts may be involved.
Your canvas
Running jobsPriorities & progress
Recovered workCheckpoints & results
The interview, drawing canvas and final report are included with Full Access.