An interactive Databricks cluster continues to consume DBUs and cloud infrastructure while it is running without useful work.
Auto-termination stops a cluster after a configured period without activity. The setting limits idle time, but it does not reduce the cost of queries that are running or queued. A longer interval can preserve a warm cache and avoid repeated startup; a shorter interval can reduce idle billing.
Measure cluster runtime, active workload time, startup latency, DBUs, and cloud infrastructure charges together. A 30-minute setting may fit an interactive team, while a scheduled job may need a different interval.
Cluster policies can require an idle-time limit and restrict oversized configurations. Databricks documents cluster termination.