GCP Professional Data Engineer Practice Question

Your analytics team schedules a 4-hour Spark ETL job every night on a Dataproc cluster with one master node and 20 n2-standard-4 workers. To keep the cluster available for occasional ad-hoc jobs, the team leaves it running the remaining 20 hours each day, resulting in high compute charges for mostly idle resources. Management asks you to reduce costs without extending the nightly ETL completion time or sacrificing ad-hoc flexibility. What should you do?

Enable Dataproc autoscaling on the existing cluster so worker nodes scale down during idle periods while keeping the master and two workers running.
Configure Cloud Composer to spin up a job-scoped Dataproc cluster each night (and for any ad-hoc submission), run the Spark job, then delete the cluster after completion.
Keep the persistent cluster but convert all worker nodes to preemptible VMs to lower the hourly rate.
Rewrite the workload as a continuous Dataflow streaming job that runs on a single, permanently provisioned regional Dataflow job.

Report Issue

Answer Description

Dataproc charges for every VM (including master nodes) as long as the cluster is running, even when it is idle. The most effective way to avoid paying for idle capacity is to replace the always-on (persistent) cluster with job-scoped, ephemeral clusters. A workflow scheduler such as Cloud Composer can create an appropriately sized cluster just before each nightly Spark job starts, run the workload, and delete the cluster immediately after it finishes. Ad-hoc users can submit their own on-demand Dataproc jobs that each spin up a short-lived cluster, ensuring resources are paid for only while they are used. Autoscaling or switching to preemptible workers on a persistent cluster still incurs charges for the master nodes and any baseline workers that remain running. Converting the batch process to a streaming pipeline would keep resources allocated all day and raise, not lower, costs.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.