Google Professional Data Engineer Exam Questions Instant Access

Question 1

Your company uses separate Google Cloud projects for development, staging, and production. Developers need edit access in the development project, read-only access in the staging project, and no access in the production project. You need an effective and manageable solution to assign and enforce these permissions according to Google-recommended practices. What should you do?

AConfigure firewall rules within each project's VPC to block traffic from unauthorized developer machines.

BCreate Google Groups for access levels, assign developers to groups, and grant the groups the appropriate IAM roles on each project.

CGrant individual developers specific IAM roles directly on each project.

DCreate separate VPC Service Controls perimeters for each project and use access levels.

Answer : B

Google Cloud's core recommendation for Identity and Access Management (IAM) scalability is to manage access using Google Groups rather than individual users.

Manageability: Instead of updating IAM policies in three different projects every time a developer joins or leaves the company, you simply add/remove them from a single Google Group (e.g., developers@company.com).

Enforcement: You apply the Editor role to the group in the Dev Project, the Viewer role in the Staging Project, and no roles in the Production Project. This centrally manages the 'Joiner-Mover-Leaver' process.

Correcting other options:

A: Firewalls control network traffic, not identity-based access to GCP resources like BigQuery or Dataflow.

C: Granting roles to individuals is prone to error, difficult to audit, and does not scale.

D: VPC Service Controls are for mitigating data exfiltration, not for basic user-level role management across environments.

'Grant roles to groups instead of individual users. It is easier to update the members of a group than to update the principals in your allow policies... Using groups helps you manage permissions at scale and simplifies auditing.' (Source: Best practices for using Google Groups)

'For a consistent security posture, use groups to represent job functions and assign roles to those groups at the appropriate level in the resource hierarchy.' (Source: IAM guide - Manage access to projects, folders, and organizations)

Question 2

Your team has created several BigQuery curated datasets containing anonymized industry benchmark data. You want to make these datasets easily discoverable and accessible for querying by external partner companies within their own Google Cloud projects. You need a secure and scalable solution. What should you do?

AGrant the roles/bigquery.dataViewer IAM role to the partner group email addresses on the datasets.

BPublish the datasets as listings within BigQuery sharing (Analytics Hub).

CCreate authorized views for each dataset and grant access to each partner.

DExport the datasets to partner-specific Cloud Storage buckets.

Answer : B

Analytics Hub (part of BigQuery sharing) is the enterprise-grade, Google-recommended way to share data across organizational boundaries.

Discoverability: Analytics Hub provides a central portal where partners can search for and 'subscribe' to datasets. Traditional IAM grants (A) do not provide a discovery interface.

Scalability: You can publish a single 'listing' and allow many partners to subscribe to it. Each subscriber gets a 'linked dataset' in their project.

Security & Governance: The publisher retains control. You can see who has subscribed and even enforce 'data egress' controls to prevent partners from copying or exporting the data while still allowing them to query it.

Correcting other options:

A: This works for a few users but lacks the metadata, searchability, and formal 'subscription' lifecycle of Analytics Hub.

C: Authorized views are for internal security within a project/org and don't solve the cross-org discovery problem.

D: Exporting data creates data silos, increases storage costs, and loses the real-time query benefits of BigQuery.

'Analytics Hub is a data exchange platform that lets you share data and insights at scale across organizational boundaries... Listings let you share data without replicating the shared data... Subscribers discover data through the exchange and create a linked dataset in their own project to query the data.' (Source: Introduction to BigQuery sharing)

Question 3

You are building a Dataflow pipeline to ingest customer feedback. Before loading to your data warehouse, you must validate email addresses and enrich unstructured comment strings with a generative AI sentiment classification. Invalid records need to be routed for manual review. How should you implement this pipeline?

AApply a ParDo transform in Dataflow to validate each element, use a RunInference transform to assign sentiment scores, and use side outputs to route valid/invalid records.

BAfter the Dataflow load completes, execute a Cloud Run function to scan for invalid entries and call Vertex AI to assign sentiment scores.

CUse Dataflow to load all data into BigQuery and execute a SQL MERGE statement to flag invalid records and BigQuery ML to assign sentiment scores.

DConfigure the data source system to pre-validate data before sending it to Dataflow and use the BigQuery ML.GENERATE_TEXT command to assign a sentiment score.

Answer : A

This scenario requires a sophisticated streaming or batch ETL pipeline involving validation, AI enrichment, and branching logic. Apache Beam (Dataflow) is the standard tool for this on Google Cloud.

Validation via ParDo: A ParDo (Parallel Do) transform is the fundamental way to perform element-wise logic in Dataflow. It can be used to run regex or validation logic on email strings for every record in the stream.

Enrichment via RunInference: For integrating Generative AI or machine learning models into a Dataflow pipeline, the RunInference transform is the Google-recommended approach. It manages model loading and optimization (batching requests) to services like Vertex AI or local models, allowing for efficient sentiment classification during the 'flight' of the data.

Routing via Side Outputs: This is a key feature of Apache Beam. While a transform usually produces one main output, Side Outputs allow a single ParDo to emit data to multiple 'p-collections.' One collection can contain valid records destined for the data warehouse, while another contains invalid records routed to a 'dead-letter' bucket or table for manual review.

Correcting other options:

B & C: These are 'post-processing' approaches. Moving invalid data into a warehouse or a secondary service after the load increases complexity and cost, and violates the requirement to validate and enrich before loading.

D: Relying on the source system for validation is often impossible in real-world data engineering where you don't control the source, and using BigQuery ML after the fact doesn't address the requirement of routing invalid records within the pipeline.

'Side outputs are a powerful feature of the Beam model that allow you to produce multiple output PCollections from a single ParDo. This is useful for routing data to different destinations based on certain criteria, such as sending malformed data to a dead-letter queue.' (Source: Apache Beam Programming Guide - Additional Outputs)

'The RunInference transform lets you perform internal and external model inference within your pipeline... It handles the complexities of using machine learning models in a distributed data processing system.' (Source: Dataflow ML - Use RunInference)

Question 4

You want to analyze hundreds of thousands of social media posts daily at the lowest cost and with the fewest steps.

You have the following requirements:

You will batch-load the posts once per day and run them through the Cloud Natural Language API.

You will extract topics and sentiment from the posts.

You must store the raw posts for archiving and reprocessing.

You will create dashboards to be shared with people both inside and outside your organization.

You need to store both the data extracted from the API to perform analysis as well as the raw social media posts for historical archiving. What should you do?

AStore the social media posts and the data extracted from the API in BigQuery.

BStore the social media posts and the data extracted from the API in Cloud SQL.

CStore the raw social media posts in Cloud Storage, and write the data extracted from the API into BigQuery.

DFeed to social media posts into the API directly from the source, and write the extracted data from the API into BigQuery.

Answer : D

Question 5

You're training a model to predict housing prices based on an available dataset with real estate properties. Your plan is to train a fully connected neural net, and you've discovered that the dataset contains latitude and longtitude of the property. Real estate professionals have told you that the location of the property is highly influential on price, so you'd like to engineer a feature that incorporates this physical dependency.

What should you do?

AProvide latitude and longtitude as input vectors to your neural net.

BCreate a numeric column from a feature cross of latitude and longtitude.

CCreate a feature cross of latitude and longtitude, bucketize at the minute level and use L1 regularization during optimization.

DCreate a feature cross of latitude and longtitude, bucketize it at the minute level and use L2 regularization during optimization.

Answer : B

Reference https://cloud.google.com/bigquery/docs/gis-dataa

To engineer a feature that incorporates the physical dependency of location on housing prices for a neural network, creating a numeric column from a feature cross of latitude and longitude is the most effective approach. Here's why option B is the best choice:

Feature Crosses:

Feature crosses combine multiple features into a single feature that captures the interaction between them. For location data, a feature cross of latitude and longitude can capture spatial dependencies that affect housing prices.

This approach allows the neural network to learn complex patterns related to geographic location more effectively than using raw latitude and longitude values.

Numerical Representation:

Converting the feature cross into a numeric column simplifies the input for the neural network and can improve the model's ability to learn from the data.

This method ensures that the model can leverage the combined information from both latitude and longitude in a meaningful way.

Model Training:

Using a numeric column for the feature cross helps in regularizing the model and prevents overfitting, which is crucial for achieving good generalization on unseen data.

Question 6

Your data science team needs to perform interactive SQL queries on large datasets stored in Apache Parquet format within a Cloud Storage bucket. The team is familiar with Apache Hive and wants to leverage existing HiveQL queries. You need to provide an environment for the team to run their interactive HiveQL queries directly against the data in Cloud Storage. You want to keep operational overhead to a minimum. What should you do?

ALoad the Parquet data into a BigQuery native table and use the BigQuery Connector for Hive to run the queries.

BInstall and configure an Apache Hadoop and Hive cluster manually on a group of Compute Engine instances.

CConfigure BigQuery with an external table definition pointing to the Parquet files.

DDeploy a Dataproc cluster with Hive services enabled.

Answer : D

The core requirement is to run interactive HiveQL queries (not just standard SQL) while keeping operational overhead low.

Dataproc: Dataproc is the managed service for running Apache Hadoop, Spark, and Hive on Google Cloud. It allows the team to use the exact HiveQL syntax and Hive tools they are familiar with. Because it is a managed service, it significantly reduces the overhead compared to manual installation on GCE.

Cloud Storage Integration: Dataproc is designed to use Cloud Storage as the underlying 'file system' (using the gs:// connector), allowing Hive to query Parquet files directly without moving them.

Correcting other options:

A: Loading data into BigQuery is an ingestion process that adds overhead and moves the data, contradicting the 'run directly against data in Cloud Storage' and 'minimum overhead' goals.

B: Manual installation on Compute Engine has the highest operational overhead.

C: While BigQuery external tables can query Parquet on GCS, BigQuery uses GoogleSQL syntax, not HiveQL. The requirement explicitly mentions the team's familiarity and existing queries in HiveQL.

'Dataproc is a fully managed and highly scalable service for running Apache Spark, Apache Flink, Presto, and Apache Hadoop/Hive... It allows you to use your existing open-source tools and workflows on Google Cloud without the management burden of a manual cluster.' (Source: Dataproc Overview)

'You can use Dataproc to run Hive queries... By default, Dataproc uses the Cloud Storage connector to allow Hive and other components to access data directly in Cloud Storage.' (Source: Using Hive on Dataproc)

Question 7

Flowlogistic's CEO wants to gain rapid insight into their customer base so his sales team can be better informed in the field. This team is not very technical, so they've purchased a visualization tool to simplify the creation of BigQuery reports. However, they've been overwhelmed by all thedata in the table, and are spending a lot of money on queries trying to find the data they need. You want to solve their problem in the most cost-effective way. What should you do?

AExport the data into a Google Sheet for virtualization.

BCreate an additional table with only the necessary columns.

CCreate a view on the table to present to the virtualization tool.

DCreate identity and access management (IAM) roles on the appropriate columns, so only they appear in a query.

Answer : C

Google Cloud Certified Professional Data Engineer Exam Questions