GCP-ADP Practice Test 12 — 15 Questions

Question 1

You are using Workflows to orchestrate a series of steps that call external HTTP APIs. One API occasionally returns a 503 error, and you want the workflow to retry that specific step with exponential backoff, but not retry on 404 errors. How should you configure the step?

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

Accepted Answer

Set the step's retry predicate to check for a 503 status code and configure a retry policy with exponential backoff, and do not include 404 in the predicate.. Workflows steps support a retry policy with a predicate that can inspect the error or response. By predicating on 503 only, the step retries with exponential backoff on server errors and fails fast on 404. Manual loops or global retries do not provide the same granular control, and Cloud Scheduler is not designed for in-workflow retries.

Answer

Configure the entire workflow to retry on any error by setting a global retry policy, and rely on the API to eventually return success for 404s.

Answer

Wrap the API call in a try/except block and use a loop with a sleep timer that doubles each iteration, manually checking for 503 before retrying.

Answer

Use a Cloud Scheduler job to invoke the API step repeatedly with increasing intervals until it succeeds, then trigger the rest of the workflow.

Question 2

You are designing a Cloud Composer 2 DAG that must process files as they arrive in a Cloud Storage bucket. You want to minimize latency and avoid unnecessary worker resource consumption. Which TWO approaches should you use? (Choose two.)

Accepted Answer

Configure the bucket to send object change notifications to a Pub/Sub topic, and use a Pub/Sub sensor in the DAG to wait for messages.. Both a Pub/Sub sensor and a Cloud Function triggered by Cloud Storage events provide event-driven reactions to new files, minimizing latency and avoiding polling. The Pub/Sub sensor uses native Airflow operators and no custom code, while the Cloud Function uses the Airflow REST API to trigger the DAG. Polling sensors or frequent scheduled runs add latency and waste resources.

Answer

Schedule the DAG to run every minute with a cron expression and have it list the bucket for new files.

Answer

Use a Cloud Storage sensor that checks for new files at a fixed interval.

Answer

Increase the number of workers in the Cloud Composer environment to ensure sensors never block other tasks.

Question 3

You are using Cloud Composer to orchestrate a daily ETL pipeline. A task that normally completes in 10 minutes occasionally runs for over 2 hours due to unexpected data volume. You want to prevent this task from indefinitely blocking the workflow and ensure it fails after 30 minutes so downstream tasks can be retried or alerted. What should you configure on the task?

Accepted Answer

Set execution_timeout to 30 minutes in the task definition.. In Cloud Composer, Airflow tasks support an execution_timeout parameter that defines the maximum runtime for a single task. When the timeout is reached, the task is terminated and marked as failed, enabling downstream tasks to proceed or alerts to fire. Configuring this on the specific task ensures that only that task is affected, preserving the rest of the DAG's execution.

Answer

Set the DAG's dagrun_timeout to 30 minutes.

Answer

Set the task's retries parameter to 0.

Answer

Set the scheduler's max_threads to a higher value.

Question 4

You are designing a Looker Studio report for a sales team. The report must allow users to filter data by region and product category, and also drill down from country to city. Which two features should you implement to meet these requirements? (Choose two.)

Accepted Answer

Enable drill-down on a geo chart by adding a hierarchy of country and city dimensions, and set the chart to allow drill-down.. To filter by region and product category, drop-down controls are the standard interactive elements. To drill down from country to city, a geo chart with a drill-down hierarchy is required. These two features together provide the necessary interactivity and detail. The other options either use less appropriate mechanisms (parameters, blends) or do not address both requirements.

Answer

Use a blend to combine region and product category data, and add a filter control for the blended data source.

Answer

Add a date range control and a dimension filter for region, and set the chart to display city-level data by default.

Answer

Create a parameter for region and product category, and use it in calculated fields to filter the data.

Question 5

You are designing a data ingestion pipeline that must handle late-arriving data in a streaming source. The data arrives with timestamps that may be up to 10 minutes behind the current time. You need to ensure that all data is processed correctly and that windowed aggregations include late data. Which Dataflow feature should you use?

Accepted Answer

Allowed lateness. Allowed lateness is the Dataflow feature that specifies how long to wait for late data after the watermark passes the end of a window. Setting it to 10 minutes ensures that data arriving up to 10 minutes late is still processed and included in windowed aggregations. Other triggers and modes do not address late data.

Answer

Watermark triggers

Answer

Discarding mode

Answer

Processing time triggers

Question 6

Your Workflows definition calls three Cloud Run services in sequence. The second service occasionally returns HTTP 503 during cold starts, and you want Workflows to retry those specific failures with exponential backoff while treating HTTP 400 responses as permanent failures. How should you configure this?

Accepted Answer

Add a retry block to the step with max_retries and a predicate that matches the 503 status code, and let 400 responses propagate as errors.. Workflows supports a retry policy on individual steps, including max_retries, an interval with exponential backoff, and a predicate written in CEL. By matching only the 503 status in the predicate, transient cold-start failures are retried while 400 responses fail immediately. Manual try/except loops, longer timeouts, and scheduler-based restarts do not provide the same targeted behavior.

Answer

Configure a Cloud Scheduler job to re-trigger the entire workflow whenever it fails.

Answer

Set the workflow-level execution timeout to a long value and rely on Cloud Run's own request retries to handle failures.

Answer

Wrap the second call in a try/except block and re-invoke it in the except handler until it succeeds.

Question 7

You are building a Dataflow pipeline that reads from Pub/Sub and writes to BigQuery. Some messages contain malformed JSON that causes the pipeline to fail repeatedly. You want the pipeline to continue processing valid messages and send the bad records to a dead-letter location for later analysis. Which Dataflow feature should you use?

Accepted Answer

A side output with a tagged PCollection for failed records. Side outputs in Apache Beam and Dataflow allow a DoFn to emit elements to additional PCollections using tags. By wrapping the JSON parsing logic in a try/catch and emitting failures to a tagged side output, the pipeline can continue processing valid messages and write the bad records to a separate sink such as Cloud Storage or BigQuery. Pub/Sub dead-letter topics, BigQueryIO retries, and logging sinks do not provide in-pipeline routing of malformed records.

Answer

A dead-letter queue configured in the Pub/Sub subscription

Answer

A Cloud Logging sink that exports error logs to Cloud Storage

Answer

A retry policy on the BigQueryIO write transform

Question 8

You are using Cloud Dataflow to process a batch pipeline that reads from Cloud Storage, transforms data, and writes to BigQuery. You notice that the pipeline is running slowly and want to optimize it. Which of the following should you do to improve performance?

Accepted Answer

Increase the number of workers and use a larger machine type.. To optimize a slow Dataflow batch pipeline, increasing the number of workers and using larger machine types can provide more parallelism and resources, directly addressing performance bottlenecks. Other options either change the pipeline type inappropriately or reduce resources, which would worsen performance.

Answer

Switch the pipeline to use streaming mode instead of batch mode.

Answer

Disable autoscaling to keep the worker count fixed.

Answer

Reduce the number of workers to lower contention.

Question 9

A financial analytics team ingests transaction events into Pub/Sub, then uses a Dataflow streaming pipeline with a GroupByKey transform to aggregate per-account totals. The team reports that late-arriving events—sometimes hours behind due to mobile connectivity—are being dropped and never counted in the aggregates. You need the pipeline to include those late events in the correct window. What should you configure?

Accepted Answer

Set a longer allowed lateness on the windowing strategy and emit late panes from the pipeline.. Allowed lateness is the Dataflow setting that governs how long window state persists after the watermark passes the window end. Extending it lets events arriving hours behind be assigned to the correct window and emitted as late panes. Pub/Sub retention, batch mode, and Reshuffle operate at different layers and do not change event-time windowing semantics.

Answer

Increase the Pub/Sub message retention duration to 7 days on the topic.

Answer

Switch the pipeline from streaming mode to batch mode with a longer maximum worker count.

Answer

Add a Reshuffle transform after the GroupByKey to redistribute late records across workers.

Question 10

A data analyst needs to create a Looker Studio report that shows sales data filtered by region. The report must allow users to select one or more regions from a drop-down menu, and the selection should update all charts on the page. Which control should be added to the report?

Accepted Answer

A filter control with the region dimension as the control field.. A filter control is designed for interactive filtering by dimension values and applies to all charts on the page that share the same data source. Using the region dimension allows multi-select and dynamically updates visualizations, directly fulfilling the requirement for a drop-down that filters all charts.

Answer

A data control that allows users to switch between different data sources.

Answer

A date range control with a custom date dimension.

Answer

A chart-level filter applied to each chart individually.

Question 11

You need to ingest data from a MySQL database into BigQuery. The database is updated continuously, and you need near-real-time replication with minimal impact on the source database. You also want to avoid writing custom code. What should you use?

Accepted Answer

Datastream with a MySQL source and BigQuery destination.. Datastream is designed for continuous CDC from MySQL to BigQuery. It reads the binlog with minimal impact on the source, provides near-real-time replication, and requires no custom code. Other options are either batch-oriented, require coding, or do not replicate data into BigQuery.

Answer

Dataflow with a MySQL to BigQuery template.

Answer

Cloud SQL federated queries from BigQuery.

Answer

BigQuery Data Transfer Service with a MySQL connector.

Question 12

Your team runs a nightly Apache Spark job on a Dataproc cluster to cleanse raw event logs stored in Cloud Storage. The job currently reads the raw data, transforms it, and writes Parquet files back to Cloud Storage for downstream BigQuery external tables. The job duration has grown from 20 minutes to over 3 hours as data volume increased, and the cluster is a fixed-size 10-node cluster. You want to reduce runtime without rewriting the Spark code. What should you do?

Accepted Answer

Enable Dataproc autoscaling on the cluster and configure a minimum and maximum worker count.. Dataproc autoscaling is the built-in mechanism for adjusting worker capacity to workload demand without modifying application code. By setting minimum and maximum worker counts, the cluster grows during shuffle-heavy phases and scales back when idle, improving runtime and controlling cost for a recurring batch job. The other choices either require rewriting code, scale the wrong node type, or change storage architecture without fixing compute capacity.

Answer

Increase the Dataproc cluster's master node machine type to a larger high-memory instance.

Answer

Move the raw event logs from Cloud Storage into BigQuery and run the Spark job against BigQuery tables.

Answer

Convert the Spark job to a Dataflow pipeline using Apache Beam and run it on Dataflow.

Question 13

You need to prepare data for analysis in BigQuery. You have a CSV file in Cloud Storage that contains personally identifiable information (PII) such as email addresses and phone numbers. You must obfuscate this PII before loading it into BigQuery. Which two Google Cloud services or features can help you achieve this? (Choose two.)

Accepted Answer

Cloud Dataflow with the DLP transform. Cloud DLP API and Dataflow with the DLP transform are both designed to inspect and de-identify sensitive data. Cloud DLP API can process data in Cloud Storage directly, while Dataflow can integrate DLP as a pipeline step. Together or separately, they can obfuscate PII before loading into BigQuery. The other options provide access control or encryption, but not obfuscation.

Answer

BigQuery row-level security

Answer

BigQuery column-level security

Answer

Cloud Storage default encryption

Question 14

You need to load a 4 TB uncompressed CSV file from Cloud Storage into BigQuery. The file has 200 columns, and you want the load to complete as quickly as possible while minimizing cost. The data is already clean and does not require transformation. Which approach should you take?

Accepted Answer

Convert the CSV to compressed Parquet, split into multiple files, and load the Parquet files into BigQuery.. For a large, clean dataset, converting to columnar compressed Parquet and splitting into multiple files gives BigQuery the best parallelism and compression, reducing both load time and cost. A direct load of Parquet avoids the overhead of parsing uncompressed CSV and does not require an external table or streaming infrastructure. This matches the goal of fast, low-cost ingestion without transformation.

Answer

Use the BigQuery Storage Write API to stream the CSV rows into a native table.

Answer

Load the CSV directly into BigQuery using the bq load command with the --autodetect flag.

Answer

Create an external table over the CSV in Cloud Storage and query it directly without loading.

Question 15

You are using Cloud Data Fusion to build a batch pipeline that reads from an on-premises Oracle database and writes to BigQuery. The pipeline must run daily and you need to ensure that only new or changed records are ingested since the last run. Which plugin feature should you configure in the source stage?

Accepted Answer

Configure the source plugin's 'Incremental' property with a timestamp column or a numeric column to track changes.. Cloud Data Fusion's database source plugins include an incremental mode that uses a designated column to track the last processed value. By choosing a timestamp or incrementing numeric column, the pipeline can request only rows newer than the stored value on each run. This avoids full-table scans and prevents duplicate ingestion, which is exactly what a daily incremental load from Oracle to BigQuery requires.

Answer

Set the source plugin's 'Number of Splits' to a high value to parallelize reads.

Answer

Enable the 'Use Connection' property and select a preconfigured Oracle connection.

Answer

Set the sink's 'Update Schema' property to true so BigQuery can merge new records.