GCP-ADP Practice Test 14 — 15 Questions

Question 1

You are troubleshooting a Dataflow job that is running slower than expected when writing data to BigQuery. You suspect the issue is related to hot keys. What is the recommended strategy to mitigate this?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Add a random salt to the key before the shuffle operation. Hot keys cause a small number of keys to be processed by a single worker, creating a bottleneck during the shuffle. Adding a random salt (e.g., appending a random suffix) to the key before the shuffle distributes those records across many workers, then a second pass removes the salt to recombine results. This is the standard Dataflow/Apache Beam mitigation for key skew.

Answer

Change the Dataflow runner to a different region

Answer

Increase the number of worker machines

Answer

Use Cloud Storage as an intermediate sink

Question 2

You want to perform data quality checks on a BigQuery table that is updated daily. How should you schedule the Dataplex task?

Accepted Answer

Configure a CRON schedule on the task. Dataplex tasks can be scheduled using a CRON expression to run automatically at specified intervals. Since the BigQuery table is updated daily, a CRON schedule ensures the data quality checks run after each update without manual intervention. This is the standard method for recurring data quality checks in Dataplex.

Answer

Manually trigger via the UI

Answer

Wait for discovery to trigger it

Answer

Use Cloud Run to trigger the task

Question 3

Your organization uses Workflows to orchestrate multi-step API calls. You need to handle transient failures in a specific HTTP request by retrying it up to 5 times with exponential backoff. Which block should you use?

Accepted Answer

A 'try' block with a 'retry' policy.. Workflows provides a try/retry block where you can specify a retry policy with max_retries and a backoff (initial delay, multiplier, max delay), which is exactly the mechanism for retrying a transient HTTP failure up to 5 times with exponential backoff. The retry policy is attached to the try block and applies to the steps inside it.

Answer

A 'loop' block with a counter increment.

Answer

An 'if' statement checking for 500 status codes.

Answer

A global 'exception' handler at the end of the workflow.

Question 4

Which TWO of the following are recommended practices when designing a Dataflow pipeline for high-throughput streaming ingestion?

Accepted Answer

Ensure the pipeline is parallelized across workers. Option B is correct because Dataflow achieves high throughput by distributing work across many worker VMs, so the pipeline must be designed to parallelize elements (for example, avoiding artificial fusion bottlenecks and ensuring sufficient sharding) so that multiple workers can process data concurrently. Option D is correct because streaming pipelines must persist state for windows, timers, and deduplication, and excessive or unnecessary state increases checkpointing, shuffling, and persistence overhead, which limits throughput and increases latency; minimizing state keeps the pipeline lean and scalable. Option A is not recommended because global windows never close, so they cannot emit results until the watermark passes the end of time, which is unsuitable for high-throughput streaming ingestion that needs timely output. Option C is not a required practice because Dataflow can write directly to BigQuery via the BigQueryIO connector, and forcing an intermediate Cloud Storage step adds latency and complexity rather than improving throughput. Option E is not recommended because not all logic belongs in ParDo; built-in transforms such as GroupByKey, Combine, and windowing/triggering primitives are more efficient and often necessary for correct streaming semantics.

Answer

Use global windows for all streaming data

Answer

Always write data to Cloud Storage before BigQuery

Answer

Use ParDo transforms for all logic

Question 5

Which THREE features does Dataprep provide?

Accepted Answer

Data profiling. Dataprep provides data profiling (A), which automatically analyzes source data to surface schemas, data types, value distributions, and quality issues so users understand their datasets before building recipes. It also provides sampling (B), letting users preview and work with a subset of large datasets to design and test transformations quickly without processing the full volume. Data transformation (E) is a core Dataprep capability, offering a visual, serverless recipe interface (plus Wrangler-style transforms) to clean, standardize, join, and reshape data at scale. Direct database writes (C) are not a Dataprep feature, since Dataprep outputs transformed data to destinations like Cloud Storage or BigQuery rather than writing directly back to source databases. Machine Learning model training (D) is not provided by Dataprep; that belongs to services such as BigQuery ML or Vertex AI, while Dataprep focuses on data preparation.

Answer

Direct database writes

Answer

Machine Learning model training

Question 6

What is the purpose of a 'Data Zone' in Dataplex?

Accepted Answer

To provide a logical grouping for governance and policies. A Data Zone in Dataplex is a logical construct that groups data assets (such as Cloud Storage buckets, BigQuery datasets, and BigLake tables) so that a single set of governance policies, discovery, and access controls can be applied across them. It does not provision or represent physical storage; it is an organizational boundary that maps to underlying resources. This lets data stewards manage metadata, quality, and security consistently without moving or duplicating data.

Answer

To execute Spark jobs

Answer

To store query results

Answer

To act as a physical storage location

Question 7

What happens if you delete a Dataplex Lake?

Accepted Answer

The lake is deleted, but the underlying assets remain. Deleting a Dataplex Lake removes the lake resource itself but does not delete the underlying data assets (such as Cloud Storage buckets or BigQuery datasets) that were attached to it. The assets remain intact and can be re-attached to another lake or managed separately. This design prevents accidental data loss and allows for flexible re-organization of data governance structures.

Answer

The data is moved to a 'trash' bucket

Answer

Access to the assets is granted to all users

Answer

All underlying data assets are deleted

Question 8

A BigQuery table has 10 billion rows. You notice a query performing a cross-join is timing out. What should you do?

Accepted Answer

Replace the cross-join with a specific JOIN condition.. A cross-join produces the Cartesian product of two tables — with 10 billion rows, even a modest second table explodes the result set and causes the query to time out. The correct fix is to replace the cross-join with a proper JOIN that includes a specific join condition (ON clause), which restricts the output to matching rows and dramatically reduces the data processed.

Answer

Use a wildcard table.

Answer

Cache the results in BigQuery.

Answer

Increase the number of slots.

Question 9

What is the primary function of the 'Dataplex Discovery' process?

Accepted Answer

To automatically catalog and profile data. Dataplex Discovery is the automated process that scans data assets (Cloud Storage buckets, BigQuery datasets) to infer schemas, extract metadata, and profile data (e.g., statistics, data quality). It populates the Dataplex Catalog so users can search, understand, and govern data without manual cataloging. Its primary function is therefore to automatically catalog and profile data.

Answer

To manage VPC firewall rules

Answer

To run SQL queries

Answer

To migrate data to the cloud

Question 10

A marketing manager needs to see daily sales trends in Looker. The underlying data is in BigQuery and is updated every 6 hours. What is the best way to visualize this?

Accepted Answer

Create a Looker Explore and build a time-series line chart.. Creating a Looker Explore and building a time-series line chart is the best approach because Looker connects directly to BigQuery, supports scheduled/real-time data refresh, and provides interactive visualization tailored to trend analysis. A time-series line chart is the appropriate visualization for daily sales trends over time, and Looker's Explore lets the marketing manager self-serve without manual data exports.

Answer

Create a static PDF report sent via email.

Answer

Export BigQuery data to a CSV file.

Answer

Build a dashboard in Cloud Monitoring.

Question 11

You need to perform a time-series analysis in BigQuery. Your data contains timestamps at a millisecond level, but you need to aggregate by hour. What is the most performant way to truncate the timestamp?

Accepted Answer

TIMESTAMP_TRUNC(timestamp_column, HOUR).. TIMESTAMP_TRUNC(timestamp_column, HOUR) is the purpose-built BigQuery function for truncating a TIMESTAMP to a specified date part (HOUR, DAY, MONTH, etc.), returning a TIMESTAMP value that can be grouped or aggregated directly. It operates natively on the TIMESTAMP type without type conversion, so it is the most performant and semantically correct option for hourly time-series aggregation.

Answer

EXTRACT(HOUR FROM timestamp_column).

Answer

CAST(timestamp_column AS STRING).

Answer

Using a CASE statement to manually subtract milliseconds.

Question 12

You need to ingest IoT data into BigQuery. You need to handle messages arriving in millions per second. Which service acts as the buffer?

Accepted Answer

Pub/Sub. Pub/Sub is Google Cloud's fully managed, globally scalable messaging service designed to decouple producers from consumers, and it can absorb millions of messages per second without backpressure. In an IoT ingestion architecture, Pub/Sub acts as the durable buffer that holds incoming telemetry until BigQuery (via Dataflow or the Storage Write API) can commit it. This decoupling prevents data loss during BigQuery load spikes and lets each side scale independently.

Answer

Cloud SQL

Answer

Cloud Functions

Answer

BigQuery Streaming API

Question 13

When setting up a new Dataplex Lake, which TWO of the following configurations are required? (Choose two)

Accepted Answer

A unique name for the Lake. Option A is correct because every Dataplex Lake must be given a unique name within its project and location; this name is the resource identifier used in the lake's fully qualified path (projects/PROJECT/locations/LOCATION/lakes/LAKE) and is required at creation time. Option C is correct because a Dataplex Lake is a regional resource, so you must specify a geographic region (e.g., us-central1) when creating it; the region determines where the lake's metadata and associated assets are stored and cannot be omitted. Option B is not required because Dataplex uses Google-managed encryption by default, and a customer-managed encryption key (CMEK) is an optional configuration rather than a mandatory one. Option D is not required because a lake is a logical container for zones and assets; it does not need an external URL, as data is referenced through assets within zones rather than a lake-level URL. Option E is not required because access is governed by IAM roles and policies, not by supplying a list of all organization users during lake creation.

Answer

A default encryption key

Answer

An external URL for the data

Answer

A list of all users in the organization

Question 14

You are running a Dataflow job and notice that it is consuming too much memory and crashing. How do you optimize this?

Accepted Answer

Select a higher memory machine type. Dataflow worker memory is determined by the machine type; if a job is OOMing, selecting a higher-memory machine type (e.g., n1-highmem or a custom machine type) gives each worker more RAM to hold state, buffers, and user code. This directly addresses the memory pressure without changing parallelism.

Answer

Increase the number of workers

Answer

Reduce the number of workers

Answer

Use Batch mode

Question 15

Which TWO of the following are valid ways to trigger a Cloud Composer DAG?

Accepted Answer

Airflow Web UI. Option C (Airflow Web UI) is correct because Cloud Composer exposes the Airflow web server, and from the DAGs view you can manually trigger a DAG run using the 'Trigger DAG' (play) button, which creates a dag_run via the Airflow metadata database. Option E (Airflow CLI) is correct because Cloud Composer supports running Airflow CLI commands such as 'gcloud composer environments run ENVIRONMENT --location LOCATION dags trigger -- DAG_ID', which invokes the same scheduler-side trigger mechanism. Option A is wrong because the Airflow scheduler does not receive triggers by email; it only schedules DAG runs based on the DAG's schedule and metadata. Option B is wrong because the Cloud Composer console environment list does not provide a 'Run' button to trigger a DAG; it only manages environments. Option D is wrong because editing the DAG file in the bucket merely updates the DAG definition and does not itself trigger a run.

Answer

Sending an email to the Airflow scheduler.

Answer

Clicking 'Run' in the Cloud Composer console environment list.

Answer

Directly editing the DAG file in the bucket.