Answer: They can institute a retry policy for the task that periodically fails

## Explanation **Correct Answer: D** - They can institute a retry policy for the task that periodically fails **Why this is correct:** 1. **Targeted approach**: The problem is specific to one task that fails only 10% of the time. Implementing a retry policy for just that task is the most efficient solution. 2. **Cost-effective**: Retrying only the failing task minimizes compute costs compared to retrying the entire job or running multiple job instances. 3. **High success probability**: With a 90% success rate, a few retries should ensure completion without excessive resource consumption. 4. **Standard practice**: Task-level retry policies are a common pattern for handling intermittent failures in data pipelines. **Why other options are incorrect:** **A. They can institute a retry policy for the entire Job** - This would retry all tasks, not just the failing one, wasting compute resources on tasks that already succeeded. - Increases costs unnecessarily by re-running successful tasks. **B. They can observe the task as it runs to try and determine why it is failing** - While investigation is important for root cause analysis, it doesn't ensure job completion each night. - This is a reactive approach rather than a proactive solution for ensuring completion. **C. They can set up the Job to run multiple times ensuring that at least one will complete** - This would significantly increase compute costs by running multiple instances of the entire job. - With a 90% success rate, running multiple instances is wasteful overkill. **E. They can utilize a Jobs cluster for each of the tasks in the Job** - This would increase costs by provisioning separate clusters for each task. - Doesn't address the intermittent failure issue - just changes the execution environment. **Best Practice Recommendation:** In Databricks Jobs, you can configure task-level retry policies with: - Maximum number of retries (e.g., 3-5) - Exponential backoff between retries - Specific error handling for known failure patterns This approach ensures job completion while maintaining cost efficiency for intermittent failures.

Author: Keng Suppaseth