A 311 production job must not reuse interactive development clusters that analysts keep busy. What isolation practice is correct?

A 311 production job must not reuse interactive development clusters that analysts keep busy; the correct isolation practice is dedicated job compute for the production Lakeflow Job so interactive notebooks cannot starve production tasks. Job compute spins for the run and is not shared with ad-hoc analyst sessions, protecting SLA for ticket transforms that feed public status sites and internal routing. Contended all-purpose clusters create noisy-neighbor failures exactly when call volume spikes.

Configuring the production Lakeflow Job to use a dedicated all-purpose cluster so interactive notebooks cannot delay production tasks fails because all-purpose clusters are designed for interactive development, not automated production runs; even when dedicated, it lacks job compute's ephemeral lifecycle and reliability guarantees for production.

Configuring the production Lakeflow Job to use dedicated SQL warehouse compute so interactive notebooks cannot starve production tasks fails because SQL warehouses are optimized for serving SQL queries and BI workloads, not for running Lakeflow Jobs; production transforms need job compute, not a warehouse endpoint.

Configuring the production Lakeflow Job to use dedicated single-node job compute so interactive notebooks cannot starve production tasks fails because single-node compute offers no distributed scale-out or resilience for a production 311 workload; production needs multi-node job compute to handle spikes reliably.

Exam caveat: production isolation = job compute (and policies), not dedicated all-purpose. Pair isolation with permissions so casual users cannot attach or terminate prod job runs. This is related to—but distinct from—cost guards: both point away from interactive for prod, for reliability and FinOps reasons.

Operational check: confirm the 311 production job uses dedicated job compute, that interactive clusters used by analysts are not referenced by the job, and that a busy interactive session in the workspace does not delay or cancel the production run. Confirm the job is not running on SQL warehouse compute and that a warehouse endpoint is not substituted for job compute for production work.