GCP-ADP Practice Test 16 — 15 Questions

Question 1

You need to ensure data compliance for PII. Which THREE of the following tools or features should you use in combination with Dataplex? (Choose three)

Accepted Answer

Cloud Audit Logs to monitor access. Cloud Audit Logs (A) is correct because it records Admin Activity, Data Access, and System Event logs for Dataplex and underlying services, providing the audit trail needed to demonstrate who accessed PII and when for compliance reporting. Cloud DLP API (B) is correct because it discovers, classifies, and de-identifies PII (e.g., via infoType detectors and masking, tokenization, or format-preserving encryption) across data scanned through Dataplex, which is essential for identifying and protecting sensitive data. Policy Tags (E) is correct because Dataplex integrates with BigQuery column-level security via Policy Tags in Data Catalog, enabling fine-grained, tag-based access control so only authorized principals can read PII columns. Cloud Build (C) is not a data encryption service—encryption is handled by Cloud KMS and default Google-managed encryption—so it does not address PII compliance. Cloud Scheduler (D) is only a cron-style job trigger and does not itself perform data deletion or enforce retention policies, so it is not the right tool for compliance here.

Answer

Cloud Build for data encryption

Answer

Cloud Scheduler for data deletion

Question 2

You are working in a Vertex AI Workbench notebook and want to list the files in your current working directory. What command should you run?

Accepted Answer

!ls. In a Vertex AI Workbench notebook (Jupyter-based), shell commands are executed by prefixing them with an exclamation mark (!). Therefore, !ls runs the Linux ls command in the notebook's underlying VM and lists the files in the current working directory. This is the standard Jupyter/IPython syntax for invoking shell commands from a code cell.

Answer

find .

Answer

dir

Answer

list_files()

Question 3

You are designing a data ingestion pipeline that uses Pub/Sub to capture real-time events. You need to ensure that messages are processed exactly once and in order. Which two features should you configure? (Choose two.)

Accepted Answer

Use exactly-once delivery on the subscription. To achieve exactly-once and ordered processing in Pub/Sub, you must enable both message ordering and exactly-once delivery on the subscription. Message ordering ensures sequence per ordering key, while exactly-once delivery prevents duplicates. Other options like dead-letter topics or ack deadlines address different concerns and do not provide these guarantees.

Answer

Increase the number of subscriber clients

Answer

Create a dead-letter topic for failed messages

Answer

Set the acknowledgment deadline to 10 minutes

Question 4

A data analyst is using BigQuery ML to create a linear regression model to predict customer lifetime value based on historical data. After training the model, they want to evaluate its performance. Which BigQuery ML function should they use to get evaluation metrics such as R-squared and mean absolute error?

Accepted Answer

ML.EVALUATE. To evaluate a BigQuery ML model, ML.EVALUATE is the appropriate function. It computes standard metrics for the model type, such as R-squared and mean absolute error for linear regression. ML.PREDICT generates predictions, ML.TRAINING_INFO shows training progress, and ML.FEATURE_INFO describes input features, none of which provide the required evaluation metrics.

Answer

ML.TRAINING_INFO

Answer

ML.PREDICT

Answer

ML.FEATURE_INFO

Question 5

A retail analytics team stores raw clickstream events in a Cloud Storage bucket. They want to query this data with BigQuery using standard SQL without loading it into BigQuery native storage, and they want to minimize cost by avoiding duplicate storage. Which Google Cloud feature should they use?

Accepted Answer

BigQuery external tables over Cloud Storage. BigQuery external tables allow querying data stored in Cloud Storage or other external sources without ingesting it into BigQuery managed storage. This satisfies both the standard SQL requirement and the cost goal of avoiding duplicate storage. Other options either copy data into BigQuery or serve entirely different purposes such as database replication or filesystem mounting.

Answer

Cloud Storage FUSE with a BigQuery remote function

Answer

BigQuery Data Transfer Service

Answer

Datastream with a BigQuery destination

Question 6

A data analyst is working in BigQuery and needs to create a visualization that shows the relationship between advertising spend and revenue for each day over the past month. The data is in a table with columns `date`, `ad_spend`, and `revenue`. They want to quickly identify if there is a correlation. Which type of chart should they use in Looker Studio?

Accepted Answer

Scatter chart. A scatter chart is designed to display the relationship between two numerical variables by plotting each data point according to its values on both axes. In this scenario, plotting ad spend on one axis and revenue on the other allows the analyst to visually assess correlation and outliers. Time series, bar, and pie charts do not serve this purpose.

Answer

Time series chart

Answer

Pie chart

Answer

Bar chart

Question 7

In a Cloud Composer DAG, you need to trigger a BigQuery job that runs as a service account other than the one assigned to the GKE nodes. How should you implement this?

Accepted Answer

Use the 'impersonation_chain' parameter in the BigQueryInsertJobOperator.. The 'impersonation_chain' parameter is the native Airflow mechanism for a task to assume a different service account identity at runtime. When set on BigQueryInsertJobOperator, the operator calls the IAM Credentials API to generate short-lived access tokens for the target service account, so the BigQuery job executes under that SA's permissions rather than the Composer worker SA. This preserves least privilege without storing any long-lived secrets.

Answer

Hardcode the credentials in a JSON file inside the DAG folder.

Answer

Change the worker node's identity in the Cloud Console.

Answer

Configure a connection with the 'Service Account Key' type in Airflow UI.

Question 8

Which TWO of the following are common Dataflow pipeline patterns?

Accepted Answer

Streaming to BigQuery. Option A (Streaming to BigQuery) is correct because a canonical Dataflow pattern is reading unbounded data from a source such as Pub/Sub and writing it into BigQuery for near-real-time analytics, typically using the BigQueryIO connector with streaming inserts or Storage Write API. Option C (Batch processing files from GCS) is correct because another standard Dataflow pattern is batch ingestion of files stored in Cloud Storage, reading them with TextIO/AvroIO/ParquetIO and transforming them before loading to a sink such as BigQuery or GCS. The remaining options are not Dataflow pipeline patterns: B (UI-based data drag-drop) describes a visual ETL tool like Cloud Data Fusion or DataPrep, not Dataflow's Apache Beam programming model; D (Direct database connection) is not a typical Dataflow pattern since Dataflow sources/sinks are Beam I/O connectors rather than direct JDBC-style database connections; and E (Cloud Function execution) is an event-driven serverless compute service, not a Dataflow pipeline pattern.

Answer

UI-based data drag-drop

Answer

Direct database connection

Answer

Cloud Function execution

Question 9

Your team runs a Cloud Composer 2 environment that processes financial data. A critical DAG must run exactly once per day at 02:00 UTC, and the Airflow scheduler must not backfill missed runs from the previous week when the DAG is unpaused. You need to configure the DAG so that only future scheduled runs execute. What should you set in the DAG definition?

Accepted Answer

Set catchup=False in the DAG constructor and define schedule='0 2 * * *'.. The requirement is a daily 02:00 UTC run with no historical backfill. In Cloud Composer, catchup=False suppresses the creation of runs for intervals that elapsed before the DAG was enabled, while the cron schedule controls future timing. Other parameters such as max_active_runs or start_date adjustments do not disable backfill, and schedule=None removes the automatic cadence entirely.

Answer

Set catchup=True and use a TimeDeltaSchedule with timedelta(days=1).

Answer

Set start_date to the current date and schedule_interval=None in the DAG constructor.

Answer

Set max_active_runs=1 and schedule='0 2 * * *' in the DAG constructor.

Question 10

You are building a Looker Studio report and need to display a calculated field that shows the profit margin as a percentage. The data source contains fields for revenue and cost. Which expression should you use for the calculated field?

Accepted Answer

(SUM(revenue) - SUM(cost)) / SUM(revenue). To calculate profit margin as a percentage, you need to aggregate revenue and cost separately, then compute (total revenue - total cost) / total revenue. This ensures the result reflects the overall profitability across all rows, not an average of per-row margins. The expression using SUM for both revenue and cost correctly implements this.

Answer

SUM(revenue - cost) / SUM(revenue)

Answer

SUM(revenue) / SUM(cost) - 1

Answer

AVG(revenue - cost) / AVG(revenue)

Question 11

A fintech firm streams transaction events into a Pub/Sub topic. The analytics team needs near-real-time aggregates in BigQuery and must also retain the raw events for seven years in Cloud Storage for compliance. They want a managed service with minimal code. Which pipeline should you recommend?

Accepted Answer

Use Dataflow with a streaming pipeline that writes to BigQuery and also writes raw events to Cloud Storage.. Dataflow is the managed stream-processing service that can read Pub/Sub, compute windowed aggregates into BigQuery, and write the raw stream to Cloud Storage in the same pipeline. That covers near-real-time analytics and multi-year retention with minimal custom operational code. The alternatives either cannot retain messages long enough, require hand-built processing, or rely on a scheduler that does not process streams.

Answer

Use a Cloud Scheduler job to pull from Pub/Sub every minute and load batches into both BigQuery and Cloud Storage.

Answer

Configure a Pub/Sub subscription with a BigQuery subscription, and enable message retention on the topic for seven years.

Answer

Run a Cloud Run service that reads from Pub/Sub, writes aggregates to BigQuery, and archives to Cloud Storage.

Question 12

A retail company stores point-of-sale events in Pub/Sub and wants them available for SQL analysis in BigQuery with minimal operational overhead and near-real-time latency. The team has no servers to manage and wants the simplest supported path. Which approach should they use?

Accepted Answer

Use the BigQuery subscription type on the Pub/Sub topic to write messages directly into a BigQuery table.. Pub/Sub provides a native BigQuery subscription type that streams messages directly into a table, eliminating the need to run or maintain any subscriber infrastructure. It is the lowest-overhead way to achieve near-real-time SQL analysis, and it handles delivery concerns such as retries and dead-lettering, which custom pipelines would have to implement manually.

Answer

Schedule a Cloud Scheduler job that runs a Dataflow batch pipeline every fifteen minutes to read from Pub/Sub and write to BigQuery.

Answer

Export Pub/Sub messages to a Cloud Storage bucket and load them into BigQuery with a scheduled load job.

Answer

Write a Cloud Run service that subscribes to the topic and calls the BigQuery Storage Write API for each message.

Question 13

A retail company ingests point-of-sale events into BigQuery. Analysts frequently filter by store_id and by transaction date, and the table grows by several hundred gigabytes per month. You want to reduce the bytes scanned by these filtered queries with the least effort. What should you do?

Accepted Answer

Partition the table by transaction date and cluster it by store_id.. Filtering by date and store on a large, growing event table is the classic case for partitioning on the date column and clustering on the secondary filter column. Partition pruning removes irrelevant date ranges, and clustering limits blocks scanned within each partition, lowering cost and improving speed. The remaining choices either accelerate only narrow query types, add a caching layer without changing storage, or merely block expensive queries.

Answer

Create a materialized view that pre-aggregates sales by store and day.

Answer

Set a maximum bytes billed limit on the dataset to force smaller scans.

Answer

Enable the BigQuery BI Engine reservation for the dataset.

Question 14

Your organization stores raw event data in a Cloud Storage bucket that receives new files throughout the day. Analysts want to query this data with SQL in BigQuery without loading it into native BigQuery storage first. The data is in Parquet format and partitioned by date in the folder structure. What should you create?

Accepted Answer

An external table in BigQuery that points to the Cloud Storage bucket and uses the Parquet format and hive-partitioned layout.. A BigQuery external table lets analysts query Parquet files in Cloud Storage directly with SQL, and hive partitioning enables partition pruning on the date folders. This avoids loading data into BigQuery storage and keeps new files immediately queryable. Load jobs copy data, Dataproc is heavier and not SQL-in-BigQuery, and Cloud SQL does not natively read Parquet from Cloud Storage.

Answer

A Cloud SQL for PostgreSQL instance with a foreign data wrapper pointing at the Cloud Storage bucket.

Answer

A BigQuery table created with a load job that reads the Parquet files from Cloud Storage.

Answer

A Dataproc cluster running Spark that reads the Parquet files and exposes them through a Hive metastore.

Question 15

You are designing a data lake in Cloud Storage and need to ensure that data is organized for efficient querying by BigQuery and other analytics services. You plan to use a Hive-style partitioned directory structure. Which two practices should you follow? (Choose two.)

Accepted Answer

Use a consistent key=value naming convention for partition directories, such as dt=2024-01-01.. Using a consistent key=value directory naming convention and columnar file formats like Parquet or ORC are best practices for a Hive-partitioned data lake. The naming convention enables automatic partition inference and pruning, while columnar formats reduce I/O and improve query efficiency. Together, they optimize both cost and performance for analytics services like BigQuery.

Answer

Store all files in a single directory and rely on file name prefixes for filtering.

Answer

Store each partition as a single large file to minimize the number of files.

Answer

Ensure that the partition key is the last element in the directory path, such as year=2024/month=01/day=01.