Upstream writes land to Cloud Storage from anywhere in the US. Dataproc Spark jobs run in us-central1. You need disaster recovery for a single-region failure with RPO ≤ 15 minutes and minimal read latency. What storage architecture meets this?

The correct option uses a dual-region bucket with turbo replication, which provides an RPO of seconds, easily meeting the 15-minute requirement. Dataproc in us-central1 reads from the nearest replica, minimizing latency. In a single-region failure, Dataproc can be redeployed to us-south1 and continue reading from the same dual-region bucket, as data is replicated across both regions. Option 1's hourly Storage Transfer Service copy violates the RPO ≤ 15 minutes. Option 2 uses a US multi-region bucket, which has higher latency for Dataproc in us-central1 than a dual-region bucket with local reads. Redeploying to us-central2 doesn't address the latency issue for reads from a multi-region bucket. Option 3 suggests reading from us-south1 while Dataproc is in us-central1, which introduces unnecessary cross-region latency during normal operations.