"Design a pipeline for daily sales reporting."
Structure your answer: sources, transformations, dependencies, failure handling, monitoring.
"A pipeline failed midway. How do you recover?"
Explain idempotent design. Show how you'd clear failed tasks and rerun from the failure point, not from scratch.
"How do you handle late-arriving data?"
Discuss reprocessing strategies. Partition by arrival time vs event time. Consider downstream impact.
Always mention monitoring and alerting. Pipelines fail. The question is how fast you detect it.