Data-Engineer-Associate - Databricks Intelligence Platform (6% of the exam) - Section 1.2

Understand the platform's compute services, including their characteristics, limitations, and cost models, and select the most suitable option for each workload.

Compare all-purpose compute, jobs compute, serverless compute, and SQL warehouses by startup time, concurrency, isolation, and billing, and choose the right option for interactive development, scheduled ETL, and ad hoc BI queries. Recognise the limitations of single-node compute and how autoscaling affects cost.

all-purpose computejobs computeserverless computeSQL warehousesautoscaling

Practice question for this objective

Free sampleDatabricks Intelligence Platformmedium

An analytics team supports a dashboard that is queried in short bursts during the working day and sits idle for hours in between. They are comparing a serverless SQL warehouse with a classic SQL warehouse. Which statement correctly describes the difference between the two?

  • AA classic warehouse resumes in a few seconds because its virtual machines are held ready in the customer's cloud account, whereas a serverless warehouse has to acquire capacity for each query it receives.
  • BA serverless warehouse draws on compute managed in the Databricks account and becomes available within seconds, whereas a classic warehouse provisions virtual machines in the customer's cloud account and takes minutes. Correct
  • CA serverless warehouse accepts read queries alone, so any statement that writes to a Unity Catalog table has to be routed to a classic warehouse running in the same workspace.
  • DThe two run on identical infrastructure and differ only in how the charge is presented on the bill, so the start-up time and the location of the compute are the same for both.
Serverless SQL warehouses start in seconds on Databricks-managed compute, while classic warehouses launch virtual machines in the customer's cloud account. The two warehouse types differ in where the compute lives. Serverless capacity is managed and kept ready in the Databricks account, so a warehouse can be started or resumed in seconds and can stop aggressively during idle periods without punishing users with a cold start. A classic warehouse must request and boot virtual machines in the customer's own cloud account first, which is why its start-up is measured in minutes and why teams tend to leave it running.

Why A is wrong: This inverts the two, which is a common mix-up; classic warehouses are the ones that must provision virtual machines on resume, and serverless warehouses do not acquire capacity separately for each query.

Why B is correct: Serverless SQL warehouses use capacity that Databricks manages and keeps ready, so a start or resume completes in seconds, while a classic warehouse must launch virtual machines in the customer's cloud subscription first.

Why C is wrong: It sounds like a reasonable restriction on managed compute, but serverless SQL warehouses run the same SQL surface as classic ones, including writes such as MERGE and INSERT.

Why D is wrong: Billing does differ, but that is a consequence rather than the distinction; the compute genuinely runs in different places and becomes available on very different timescales.

See more Data-Engineer-Associate practice questions, answers explained.

Exam traps in Databricks Intelligence Platform

Answers that look right on this material and are not. Each one is a distractor from a different question in the Data-Engineer-Associate bank for this domain.

  • An all-purpose cluster with auto termination disabled, so that the compute remains attached to the notebook throughout the day and the engineer never waits for instances again.

    Why it is wrong: Removing auto termination does remove the startup wait, but the cluster is then billed for every idle pause as well, which directly contradicts the requirement that the pauses cost nothing.

  • The cluster stays running between the nightly runs so that the next run starts immediately, and the workspace is charged the all-purpose rate for the idle hours in between.

    Why it is wrong: This describes the behaviour of a long-lived all-purpose cluster, which is exactly what the change moves away from; jobs compute is created for the run and terminated afterwards, so no idle hours are billed.

  • Keep the existing cluster and enable autoscaling on it, so that workers are released between the nightly runs and the job pays for fewer nodes overall.

    Why it is wrong: Autoscaling can trim worker count during light periods, but it does not change the DBU rate applied to all-purpose compute, and the team would still be billed at the interactive rate for every hour the cluster is up.

Examworthy is not affiliated with or endorsed by Databricks. Original, blueprint-aligned practice material only.