DP-100 Question #318: Real Exam Question with Answer & Explanation

Question

You have an Azure Machine Learning workspace and a serverless Spark compute resource. You plan to run the same Spark session every 40 minutes. Each session runs for 15 minutes. During testing, you observe a 15-minute delay at the start of the Spark sessions. You need to reduce the delay to less than 1 minute. What should you do?

Options

• AConfigure the session timeout to be 25 minutes.
• BIncrease the number of nodes to 16.
• CEnable dynamically allocated executors.
• DEnable an isolated compute.

Explanation

{"question_number": 10, "correct_answer": "A", "explanation": "You should configure the session timeout to 25 minutes. Each Spark session runs for 15 minutes and the next session starts every 40 minutes, leaving a 25-minute idle gap between sessions (40 − 15 = 25). The 15-minute cold start delay occurs because the serverless Spark compute shuts down between sessions, requiring a full cluster bootstrap at the next run. By setting the session timeout to 25 minutes, the Spark session remains alive during the entire idle gap. When the next scheduled execution begins, it reuses the already-warm session, reducing the startup delay to under 1 minute. Increasing nodes (B) addresses throughput, not cold start. Dynamic executors (C) manages resource scaling within a session, not startup time. Isolated compute (D) does not prevent cold starts.", "generated_by": "claude-sonnet", "llm_judge_score": 4}

No community discussion yet for this question.