Job clusters are optimized for automated production workloads. They start when a job is triggered and terminate automatically once the task completes. This ensures cost control while maintaining performance and reliability for batch ETL. Autoscaling allows Databricks to add or remove workers dynamically based on workload size, ensuring large data volumes are processed efficiently.
All-purpose clusters are intended for development or ad-hoc workloads, not scheduled ETL.
Reference Source: Databricks Compute and Job Cluster Configuration Documentation - "Autoscaling and Job Clusters."