Library analytics leaves all-purpose clusters running after analysts disconnect, driving idle charges. Which setting should they apply?

Library analytics leaving all-purpose clusters running after analysts disconnect should configure automatic termination after a defined idle period so forgotten clusters stop billing. Auto-termination is the primary cost-control for interactive civic clusters: when notebook activity ceases for the configured idle window, Databricks terminates the cluster and halts VM charges, which directly addresses idle research sessions that outlast the analyst’s workday.

Disabling termination so clusters never stop maximizes the idle-charge problem the stem describes. Converting every exploratory cluster into a continuous Structured Streaming job is a category error—streaming jobs are long-running by design and would increase always-on cost, not reduce forgotten interactive clusters. Granting all library volunteers unrestricted cluster-create rights without idle policy multiplies unmanaged clusters and worsens spend and security posture.

Exam caveat: auto-termination applies to interactive/all-purpose clusters; job clusters typically end when the job ends, and streaming jobs are intentionally continuous—do not “fix” streaming by enabling notebook-style idle termination without understanding the workload. Also note that very short idle timeouts can annoy analysts; pick a policy-aligned period (for example 30–120 minutes) rather than zero or infinite.

Operational check: set a workspace/cluster-policy default auto-termination for library all-purpose clusters, audit running clusters for owners with no recent commands, and verify in the account console that terminated clusters no longer accrue DBU/VM charges after the idle window.