A weekly Dataproc Spark job runs ~30 minutes on a 15-node cluster and writes results to BigQuery. How should you optimize cluster cost given this profile?

Using preemptible virtual machines (PVMs) for the Dataproc cluster is the most cost-effective optimization. PVMs offer significant cost savings (up to 80% less than standard VMs) and are suitable for fault-tolerant workloads like Spark jobs. Since this job runs weekly for only 30 minutes, the risk of preemption is low, and even if a node is preempted, Spark can recover and continue processing. Migrating to Cloud Dataflow might be an option for long-term architectural changes but is not a direct cost optimization for the existing Dataproc job. Using higher-memory nodes or attaching SSDs would likely increase costs rather than decrease them, as they are premium resources. These options might improve performance, but the question specifically asks for cost optimization given the job's profile.