Professional Cloud ArchitectAnalyze and optimize technical and business processesMedium

A large retail company is migrating its on-premises data warehouse to Google Cloud BigQuery. They have petabytes of historical sales data stored in various formats (CSV, JSON, XML) across multiple on-premises databases and file systems. The data needs to be cleaned, transformed, and loaded into BigQuery on a daily basis. The company requires a fully managed, serverless solution that can handle large volumes of data and complex transformations without requiring extensive operational overhead. Which Google Cloud service should they use for this ETL process?

  1. ACloud Dataproc
  2. BA custom Compute Engine instance running Apache Spark
  3. CCloud SQL with federated queries to Cloud Storage
  4. DCloud Dataflow
Show answer & explanation

Correct answer: D. Cloud Dataflow

Cloud Dataflow is a fully managed, serverless service for executing Apache Beam pipelines, making it ideal for large-scale data processing, ETL, and complex transformations without operational overhead. It can handle diverse data formats and integrate with BigQuery.

Why the other options are wrong

  • A. Cloud Dataproc is a managed Apache Hadoop and Spark service, but it's not fully serverless and requires more operational management than Dataflow.
  • B. Running a custom Spark cluster on Compute Engine involves significant operational overhead and is not a fully managed, serverless solution.
  • C. Cloud SQL is a relational database and federated queries are for querying external data, not for complex, large-scale ETL processes and transformations into BigQuery.

Cloud Dataflow

Cloud Dataflow is a fully managed, serverless service for executing Apache Beam pipelines, enabling reliable and expressive data processing for ETL, batch, and stream analytics.

  • Fully managed and serverless, reducing operational overhead.
  • Supports both batch and stream processing with Apache Beam.
  • Scales automatically to handle large datasets and complex transformations.

Memory trick: Extract, Transform, Load: Data's Journey.

More Analyze and optimize technical and business processes questions