PDE Practice Test 38 — 15 Questions

Question 1

A logistics company streams vehicle telemetry into a Pub/Sub topic. Each message contains a vehicle ID and a timestamp, and the Dataflow pipeline computes per-vehicle distance using a stateful DoFn with a ValueState timer set to fire after 5 minutes of event-time inactivity. During a regional network outage, some vehicles stop sending data for 20 minutes and then resume with correctly ordered timestamps. After the outage, operators notice that some late-arriving records are being dropped before the stateful computation. Which pipeline setting should be adjusted to retain those records for processing?

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

Accepted Answer

Set an allowed lateness on the windowing strategy that exceeds the 20-minute outage gap.. Late data handling is governed by allowed lateness on the window, not by Pub/Sub delivery settings or the timer domain. When the watermark advances past a window end after a 20-minute gap, records with earlier timestamps are dropped unless the window is kept alive. Setting allowed lateness beyond the outage gap lets the stateful computation include the resumed telemetry in the correct event-time window.

Answer

Switch the pipeline's windowing from event-time to processing-time windows.

Answer

Increase the Pub/Sub subscription acknowledgement deadline so messages remain outstanding longer.

Answer

Change the stateful DoFn timer from event-time to processing-time so it fires only when data resumes.

Question 2

Which THREE metrics should be monitored for a deployed machine learning model in production?

Accepted Answer

Prediction error rate. Prediction error rate (Option B) is a direct measure of model accuracy in production, reflecting how often the model's predictions deviate from actual outcomes. Monitoring this metric is essential for detecting model degradation, data quality issues, or concept drift that can silently reduce model performance over time.

Answer

Number of replicas

Answer

Training time

Question 3

After deploying a model to Vertex AI Endpoints, the prediction responses include unexpected data. The model returns logits instead of probabilities. What is the most likely cause?

Accepted Answer

The model output is not post-processed. The most likely cause is that the model output is not post-processed. In Vertex AI Endpoints, models often output raw logits (unnormalized scores) from the final layer, and a softmax or sigmoid activation must be applied as a post-processing step to convert these logits into probabilities. Without this post-processing, the endpoint returns the raw logits, which is why the prediction responses contain unexpected data.

Answer

The model was trained with different loss

Answer

The input data is scaled incorrectly

Answer

The endpoint is not properly configured

Question 4

A user named Charlie needs to deploy a model to a Vertex AI Endpoint and also create training jobs. Which role should be assigned to Charlie?

Accepted Answer

roles/aiplatform.user. Charlie needs to deploy a model to a Vertex AI Endpoint and create training jobs. The `roles/aiplatform.user` role grants the necessary permissions to use all Vertex AI resources, including creating and managing endpoints, training jobs, models, and predictions. This role is the minimum required for a user to interact with Vertex AI services without granting broader project-level permissions.

Answer

roles/owner

Answer

roles/aiplatform.modelUser

Answer

roles/editor

Question 5

Your company runs a Dataflow streaming pipeline that processes user activity from Pub/Sub and writes aggregated results to BigQuery. Lately, the pipeline is experiencing high latency and backlog growth during peak hours. You need to troubleshoot and improve performance. Which THREE actions should you take? (Choose 3.)

Accepted Answer

Change the worker machine type to a higher CPU/memory configuration. Option A is correct because moving to a higher CPU/memory worker machine type gives each worker more compute and memory to process elements, which directly addresses CPU or memory saturation that causes high latency and backlog growth during peak hours. Option C is correct because enabling Dataflow Streaming Engine moves pipeline state and windowing execution off the worker VMs to the Dataflow service, reducing worker CPU/memory pressure and improving streaming performance and autoscaling responsiveness. Option D is correct because increasing the number of workers adds horizontal processing capacity, allowing the pipeline to drain the Pub/Sub backlog faster and keep up with peak throughput. Option B is not appropriate because shortening the window duration does not reduce the total volume of data processed and can actually increase per-window overhead and output frequency, potentially worsening latency. Option E is not appropriate because adding more subscriptions to the same topic does not increase the throughput of the existing pipeline; each subscription would need its own reader/pipeline, and extra subscriptions only add fan-out consumers rather than improving the current pipeline's performance.

Answer

Decrease the window duration to reduce data per window

Answer

Add additional Pub/Sub subscriptions to the same topic

Question 6

You manage a large-scale machine learning system that recommends products to users. The model is a deep neural network trained on TensorFlow and deployed on Vertex AI Endpoint with global load balancing. The model receives over 10,000 requests per second. Recently, the team added a new feature: the user's current geographic location (latitude/longitude). After deploying the updated model, you notice that the average prediction latency has doubled, and the error rate has increased, particularly for requests from regions far from the model's primary training data (North America). You suspect the location feature is causing issues. What should you do to diagnose and mitigate the problem?

Accepted Answer

Examine the latency breakdown using Cloud Monitoring to see if the location feature is causing computationally expensive operations, then consider feature engineering like bucketing coordinates.. The correct first step is to gather diagnostic data before making changes. Cloud Monitoring provides latency breakdowns per operation, letting you confirm whether the new location feature (e.g., raw latitude/longitude floats feeding into dense layers or distance computations) is the bottleneck. Once confirmed, feature engineering such as geohash bucketing or embedding coordinates reduces computational cost and improves generalization for out-of-region requests.

Answer

Remove the location feature from the model and retrain without it to restore performance.

Answer

Increase the number of replicas for the endpoint to handle the increased latency.

Answer

Switch to a regional endpoint in North America to reduce latency for the majority of users.

Question 7

A data engineer is monitoring a Dataflow streaming pipeline and notices that the 'System Lag' metric is increasing. Which TWO actions should be taken to diagnose the issue?

Accepted Answer

Check the Dataflow monitoring UI for each stage's throughput and backlog.. The Dataflow monitoring UI provides per-stage metrics such as throughput and backlog, which directly indicate where data is accumulating. By examining these metrics, you can identify the specific stage causing the increasing system lag, enabling targeted troubleshooting without unnecessary pipeline changes.

Answer

Cancel the pipeline and restart with a larger initial worker count.

Answer

Increase the maximum number of workers to handle backlog.

Answer

Increase the BigQuery quota for streaming inserts.

Question 8

You manage several Cloud Composer 2 environments that run production DAGs. You must define an alerting strategy that detects when a DAG run fails and when a task is stuck retrying for an unusually long time, using Cloud Monitoring. (Choose two.)

Accepted Answer

Create a Monitoring alerting policy on the airflow task instance duration or retry-related metric exposed through Cloud Monitoring for the environment.. Composer surfaces Airflow execution data in two complementary ways: structured log entries in Cloud Logging and Airflow metrics in Cloud Monitoring. A log-based alerting policy catches the explicit DAG run failure event, while a metric-based policy on task duration or retries catches a task that is hanging or looping. Together they cover both the hard failure and the stuck-retrying condition, whereas health, uptime, and audit signals describe infrastructure and control-plane state rather than workload outcomes.

Answer

Create an uptime check against the Airflow web server URL and alert when the HTTP response is not 200.

Answer

Enable Cloud Audit Logs for the Composer API and alert when any environment update method is called.

Answer

Create a Monitoring alerting policy on the composer.googleapis.com/environment/healthy metric and notify when it drops below the threshold.

Question 9

You are building a forecasting model to predict daily sales for the next 90 days using historical sales data with clear seasonality and trend. You want to use BigQuery ML with minimal manual tuning. Which model type should you choose?

Accepted Answer

ARIMA_PLUS. ARIMA_PLUS is the correct choice because it is a Google-developed, automated time-series model in BigQuery ML that handles seasonality, trend, holidays, and outliers with minimal manual tuning. It automatically performs preprocessing such as missing value imputation, holiday effect detection, and seasonality decomposition, making it ideal for forecasting daily sales over a 90-day horizon. Unlike standard ARIMA, ARIMA_PLUS requires no manual specification of p, d, q parameters or seasonal orders, aligning with the 'minimal manual tuning' requirement.

Answer

ARIMA

Answer

Boosted tree (XGBoost)

Answer

Linear regression

Question 10

Your company is building a real-time anomaly detection system for financial transactions. The system must process streams of transactions and flag anomalies within seconds. The volume is moderate (5000 transactions per second). You want a fully managed solution that integrates with BigQuery for historical analysis. Which service should you use for stream processing?

Accepted Answer

Cloud Dataflow. Cloud Dataflow is a fully managed, serverless stream and batch processing service based on Apache Beam. It natively supports real-time streaming with low latency (sub-second to seconds), integrates seamlessly with Pub/Sub for ingestion and BigQuery for output, and auto-scales to handle moderate throughput like 5000 transactions per second. Its managed nature eliminates cluster operations, making it ideal for this use case.

Answer

Cloud Pub/Sub with push subscriptions

Answer

Cloud Dataproc with Spark Streaming

Answer

Cloud Data Fusion

Question 11

A company is building a continuous training pipeline that retrains a model daily using new data from a feature store. The training data must include features computed up to the timestamp of each training run. Which architecture should be used to ensure time-consistent feature values without label leakage?

Accepted Answer

Use Vertex AI Feature Store with point-in-time lookup enabled to retrieve features as of the training timestamp.. Vertex AI Feature Store's point-in-time lookup retrieves the exact feature values as they existed at the specified training timestamp, ensuring time-consistency and preventing label leakage. This mechanism avoids using future data that would not have been available at the time of prediction, which is critical for realistic model evaluation and production performance.

Answer

Train on a fixed window of the most recent features without considering timestamps.

Answer

Store all features in a Cloud SQL database and perform a join at training time.

Answer

Use Pub/Sub to stream new features into Cloud Storage and train on the latest snapshot.

Question 12

A data engineer needs to store quarterly financial data that must remain immutable for 7 years to meet regulatory compliance. The data is accessed infrequently after the first year. Which Cloud Storage feature should be used to enforce immutability?

Accepted Answer

Object Lock with Retention Policy. Object Lock with a Retention Policy enforces immutability by preventing object deletion or modification for a specified duration. For regulatory compliance requiring 7-year immutable storage, a retention policy configured with a retention period of 7 years ensures that objects cannot be overwritten or deleted, even by the root account. This directly meets the requirement for data that must remain unchanged for the full 7-year period.

Answer

Object Lifecycle Rules

Answer

Object Versioning

Answer

IAM Conditions

Question 13

A company wants to migrate 500 TB of on-premises archival data to Cloud Storage. The data is stored on a SAN and the network link is limited to 1 Gbps. The migration must complete within 10 days. What is the MOST cost-effective approach?

Accepted Answer

Order a Transfer Appliance, copy data locally, and ship it to Google for ingestion.. The Transfer Appliance is designed for large-scale data migrations where network bandwidth is insufficient. With 500 TB at 1 Gbps, the theoretical transfer time is over 46 days, far exceeding the 10-day window. The appliance allows you to physically ship the data, bypassing network constraints entirely, making it the most cost-effective and timely solution.

Answer

Set up a Cloud VPN and use rsync over the encrypted connection.

Answer

Use BigQuery Data Transfer Service to load the data directly into BigQuery.

Answer

Use Storage Transfer Service to copy data from on-premises to GCS over the existing network.

Question 14

A company trains a custom model using TensorFlow and wants to deploy it to Vertex AI for low-latency predictions. The model is large (2 GB). Which deployment option should they choose?

Accepted Answer

Deploy to Vertex AI Endpoint with a custom container. Deploying a large (2 GB) model to Vertex AI Endpoint with a custom container allows you to package the model, its dependencies, and a serving framework (e.g., TensorFlow Serving) into a Docker image. This approach supports low-latency predictions by keeping the model loaded in memory across requests, and it can scale to handle real-time inference traffic, unlike batch or serverless options that have cold-start or size limitations.

Answer

Use Vertex AI Batch Prediction job

Answer

Deploy as a Cloud Function

Answer

Deploy to Cloud Run with minimum instances

Question 15

A company is using Pub/Sub to ingest clickstream events and Dataflow to write to BigQuery. They observe that some events are malformed and cause the pipeline to fail. They need a solution that captures malformed events without blocking the pipeline and allows reprocessing later. Which Dataflow pattern should they implement?

Accepted Answer

Write malformed events to a dead letter sink (e.g., another Pub/Sub topic or GCS bucket) and continue processing healthy events. The dead letter sink pattern is the canonical Dataflow approach for handling malformed or unparseable records: instead of throwing an exception that fails the bundle (and potentially the whole pipeline), the transform routes the bad record to a secondary sink such as a Pub/Sub topic or GCS bucket while healthy events continue downstream. This preserves pipeline throughput and keeps the malformed payloads available for later inspection, correction, and reprocessing. It is the standard implementation of the dead-letter queue pattern in Apache Beam.

Answer

Use a side input to filter malformed events before the main pipeline

Answer

Use the Reshuffle transform to reattempt failures

Answer

Use logging alerts to notify the team and stop the pipeline on error