Google Cloud Associate Data Practitioner Practice Exam

What the Google Cloud Associate Data Practitioner exam covers

• Data Preparation and Ingestion219 questions
• Data Analysis and Presentation195 questions
• Data Pipeline Orchestration131 questions
• Data Management182 questions

Free Google Cloud Associate Data Practitioner sample questions

A sample of 10 questions with answers and explanations. Sign up free to practice all 727.

1. Question 1Data Preparation and IngestionSelect all that apply

   A CSV load job fails because incoming files contain an extra trailing column that is not defined in the destination table's schema. Which two changes would allow the load job to succeed? (Choose two.)

   • AEnable allow jagged rows so BigQuery accepts rows with missing trailing values
   • BReduce the number of files loaded per job so fewer rows are scanned
   • CSwitch the write disposition from WRITE_APPEND to WRITE_TRUNCATE
   • DAdd the extra column to the destination table's schema before loadingCorrect
   • ESet the option that accepts and discards values not present in the schemaCorrect
   ✓ Correct answer: D, E

   Enabling the option that ignores values not present in the schema lets BigQuery drop the extra column while loading the rest of the row, and updating the destination schema to include the new column captures it instead. Both directly address the mismatch, unlike the other listed changes.

   Why the other options are wrong

   • AAllow jagged rows addresses rows with fewer values than expected, not an extra unexpected column.
   • BLoading fewer files per job does not change whether the schema mismatch occurs.
   • CChanging the write disposition affects how rows are written, not whether extra columns are tolerated.
2. Question 2Data Preparation and Ingestion

   A data platform team enables schema validation on a Pub/Sub topic used by several publishing teams. What benefit does this most directly provide?

   • AIt automatically compresses message payloads to cut network usage
   • BIt extends the topic's default message retention period further
   • CIt guarantees messages are delivered to subscribers in publish order
   • DIt blocks malformed messages from being accepted at publish timeCorrect
   ✓ Correct answer: D

   By enforcing a schema, Pub/Sub checks each publish request's format against the schema definition and rejects mismatched messages immediately, which keeps malformed data from ever reaching downstream consumers. It has no effect on delivery ordering, payload compression, or retention duration, which are controlled by separate settings.

   Why the other options are wrong

   • ASchema settings validate structure; they do not perform payload compression.
   • BRetention period is a separate setting unrelated to schema enforcement.
   • CDelivery ordering is controlled by ordering keys, not by schema validation.
3. Question 3Data Preparation and IngestionSelect all that apply

   A team designs a subscriber for a Pub/Sub subscription using the default at-least-once delivery guarantee. Which two statements correctly describe why they should make message processing idempotent? (Choose two.)

   • APub/Sub guarantees exactly-once processing automatically for every subscription
   • BOrdering keys eliminate the need to ever consider duplicate messages
   • CPub/Sub may redeliver a message the subscriber already processed and acknowledgedCorrect
   • DNetwork retries or delayed acknowledgments can cause the same message to arrive twiceCorrect
   • EIdempotent processing is only relevant when using push subscriptions
   ✓ Correct answer: C, D

   Because Pub/Sub retries delivery when an acknowledgment is not received in time or is delayed by network issues, the same message can reach a subscriber more than once even after it was already handled, which is exactly what idempotent processing guards against. Exactly-once behavior is not automatic on every subscription, the concern applies to both pull and push delivery, and ordering keys address sequence, not duplication.

   Why the other options are wrong

   • AExactly-once delivery is an optional setting, not the automatic default behavior of every subscription.
   • BOrdering keys control message sequence and do not prevent duplicate deliveries from occurring.
   • EDuplicate delivery can happen with pull subscriptions just as it can with push subscriptions.
4. Question 4Data Preparation and Ingestion

   A ride-sharing app needs to update a live dashboard with trip events within a few seconds of each event occurring. Which ingestion approach best fits this latency requirement?

   • AStreaming ingestion through Pub/Sub into a live processing pipelineCorrect
   • BA manual CSV file upload performed by an analyst each single morning
   • CA weekly export of trip data run as a scheduled batch job process
   • DA nightly batch load job that copies over the entire day's trip logs
   ✓ Correct answer: A

   Pub/Sub combined with a streaming consumer, such as a Dataflow job or BigQuery streaming writes, moves each event through the pipeline as it arrives, achieving the near real-time latency the dashboard needs. Nightly, weekly, or manual batch approaches all introduce delays measured in hours or days, which does not satisfy a few-second requirement.

   Why the other options are wrong

   • BA manual daily upload introduces both delay and human dependency unsuitable for this need.
   • CA weekly export is far too infrequent for a dashboard that needs near real-time updates.
   • DA nightly batch job introduces hours of delay, far more than the few seconds required.
5. Question 5Data Analysis and Presentation

   A subscription service wants to predict whether each customer will cancel their subscription in the next 30 days, labeled as yes or no in historical data. Which model type should the analyst choose?

   • AARIMA_PLUS, because it forecasts a value along a time axis
   • BLOGISTIC_REG, because it predicts a binary categorical outcomeCorrect
   • CLINEAR_REG, because cancellation is measured as a continuous score
   • DKMEANS, because it groups customers into similar risk segments
   ✓ Correct answer: B

   LOGISTIC_REG in BigQuery ML is designed for classification tasks where the label is categorical, such as yes or no for cancellation, and it outputs a predicted class along with a probability, which fits this churn scenario directly. LINEAR_REG is meant for continuous numeric labels, KMEANS only groups similar customers without a labeled outcome, and ARIMA_PLUS forecasts values over time rather than classifying a snapshot of customer attributes.

   Why the other options are wrong

   • AARIMA_PLUS forecasts a sequence over time rather than classifying an individual customer record.
   • CCancellation here is a yes or no label, not a continuous numeric score, so LINEAR_REG does not fit.
   • DKMEANS only clusters customers by similarity and does not produce a yes or no prediction.
6. Question 6Data Analysis and Presentation

   An analyst is comparing BigQuery ML versus a custom built model from a dedicated data science team for a new project. Which scenario clearly favors the dedicated data science team?

   • AThe project timeline favors a quick model built with familiar SQL skills
   • BThe label and structured features already live in one BigQuery table
   • CThe team needs a specialized architecture not offered by BigQuery MLCorrect
   • DThe analyst wants to avoid exporting data out of BigQuery for training
   ✓ Correct answer: C

   BigQuery ML covers common model types like linear regression, logistic regression, k-means, and time series forecasting, but a project requiring a custom or specialized architecture outside these built in options needs the broader tooling and expertise of a dedicated data science team. The other listed scenarios all describe conditions that favor using BigQuery ML rather than a dedicated team.

   Why the other options are wrong

   • AA quick model built with existing SQL skills is a strength of BigQuery ML, not a reason to hand the project off.
   • BHaving the label and features already together in BigQuery favors using BigQuery ML directly, not a separate team.
   • DWanting to avoid exporting data for training is a reason to use BigQuery ML, not to bring in a separate team.
7. Question 7Data Pipeline Orchestration

   A team needs a pipeline step to run as frequently as once every minute to keep a near-real-time dashboard current. Is Cloud Scheduler able to support that frequency?

   • ANo, Cloud Scheduler jobs can only ever run once per calendar day
   • BNo, the shortest interval Cloud Scheduler supports is once an hour
   • CYes, but only when the target is a Compute Engine instance itself
   • DYes, cron expressions can be set to fire as often as every minuteCorrect
   ✓ Correct answer: D

   Because Cloud Scheduler uses standard unix-cron expressions, a schedule such as '* * * * *' fires the target every minute, making it suitable for near-real-time recurring tasks. This flexibility lets teams pick a cadence anywhere from once a minute up to less frequent options like monthly.

   Why the other options are wrong

   • ARestricting jobs to once per day would rule out the frequent cadences unix-cron expressions actually support.
   • BAn hourly minimum would be far too coarse; Cloud Scheduler supports much finer-grained schedules than that.
   • CThe target type, such as HTTP or Pub/Sub, does not need to be a Compute Engine instance to run every minute.
8. Question 8Data Pipeline Orchestration

   A Cloud Run function loads new rows into a fact table, but an Eventarc redelivery risks inserting the same rows twice. Which SQL approach in BigQuery helps the function stay idempotent when this happens?

   • AA SELECT statement that only reads data without writing anything
   • BA MERGE statement that upserts rows using a unique identifierCorrect
   • CA DROP TABLE statement executed before every single load
   • DA plain INSERT statement that appends every batch, no checks
   ✓ Correct answer: B

   A MERGE statement can match incoming rows against existing rows by a natural key, updating matches and inserting only genuinely new rows, so if the same batch is loaded again after a redelivered event, no duplicate rows are created. A plain INSERT has no such matching logic and would add a second copy of every row on a repeat run.

   Why the other options are wrong

   • AA read-only SELECT does not load any new data into the fact table at all.
   • CDropping the table before every load would destroy previously loaded data, which is not the intended idempotent behavior.
   • DA plain INSERT has no deduplication logic and would create duplicate rows if the same event is redelivered.
9. Question 9Data Management

   An online retailer requires that if its order database fails, normal service must be restored within 20 minutes. Which term describes this requirement?

   • AService Level Agreement (SLA)
   • BRecovery Time Objective (RTO)Correct
   • CRecovery Point Objective (RPO)
   • DMean Time To Repair (MTTR)
   ✓ Correct answer: B

   The recovery time objective defines the maximum acceptable duration between a failure occurring and normal service being restored, guiding decisions such as standby capacity and automation of failover. It answers a question about downtime, not about the amount of data that may be lost.

   Why the other options are wrong

   • AA service level agreement is a broader contractual commitment, not specifically the internal restoration time target.
   • CRecovery point objective measures acceptable data loss in time, not how long restoration takes.
   • DMean time to repair is an operational reliability metric averaged over incidents, not a stated recovery target for this event.
10. Question 10Data ManagementSelect all that apply

    A team is choosing a Cloud Storage class for compliance backups that will almost never be read. Which two storage classes are specifically designed for this kind of rarely accessed, long-term archival data? (Choose two.)

    • AAutoclass, a feature rather than a storage class itself
    • BStandard, aimed at data that is accessed frequently
    • CArchive, aimed at data accessed about once a year or lessCorrect
    • DMulti-Regional, a legacy class name for frequently accessed data
    • EColdline, aimed at data accessed roughly once per quarterCorrect
    ✓ Correct answer: C, E

    Coldline targets data accessed roughly once every 90 days and Archive targets data accessed about once a year or less, both at low per-GB prices with higher retrieval costs, making them the right fit for compliance backups. Standard is priced for frequent access, Multi-Regional is a legacy naming concept rather than a current archival class, and Autoclass is an automation feature, not a storage class.

    Why the other options are wrong

    • AAutoclass automatically manages transitions between classes; it is not itself one of the archival storage classes.
    • BStandard is priced for frequent, active access and is far costlier than needed for rarely read backups.
    • DMulti-Regional describes a legacy location type for actively used data, not a class intended for archival.

Related Google resources

• Google Cloud Associate Data Practitioner study guideKey concepts
• Google practice examsAll Google
• Certification pathWhere this fits
• Certification exam guides & tipsBlog
• Plans & pricingFree & paid
• Google Cloud Professional Cloud Security Engineer practice examRelated
• Google Cloud Professional Cloud Network Engineer practice examRelated
• Associate Google Workspace Administrator practice examRelated

Google Cloud Associate Data Practitioner practice exam FAQ