Google Cloud Platform - Associate Data Practitioner

Free GCP - Associate Data Practitioner Practice Questions

A handful of real practice questions from our GCP - Associate Data Practitioner bank — to give you a true feel for the style and difficulty before you sign up.

1. Joining fact_orders to dim_customers in an explore, a LookML developer must avoid revenue fanout. Each order references one customer, and a customer can place many orders. Which relationship value fits?

   • A.one_to_many
   • B.many_to_many
   • C.one_to_one
   • D.many_to_one✓ Correct

   Why: A fact joined to a dimension where many fact rows map to one dimension row is a many_to_one relationship. Looker uses the relationship parameter to apply symmetric aggregates that prevent revenue from being counted multiple times across joined rows. Setting one_to_many describes the inverse direction and triggers over-counting on the order side. The one_to_one option is the most tempting distractor when the developer assumes each order has a unique customer, but it ignores that customers participate in many orders.

2. Topic schemas evolve when a new nullable field is added. Which destination schema action keeps an existing BigQuery subscription functional?

   • A.Drop and recreate the destination table with the additional column included
   • B.Add the matching nullable column to the destination BigQuery table✓ Correct
   • C.Recreate the subscription with a different name and the new schema
   • D.Disable Pub/Sub schema validation entirely on the source topic

   Why: When a topic schema adds a nullable field, the BigQuery subscription keeps writing because the destination table accepts the new column once it exists as a nullable target. Adding the column is the minimal compatible change. Dropping and recreating the table loses all historical data and forces a backfill. Recreating the subscription is unnecessary because no subscription field changes, and disabling schema validation removes the safety net that catches incompatible future revisions on the topic.

3. Finance teams refresh one materialized aggregate every six hours. The lead must receive email notification if a refresh fails. No downstream service consumes the result outside BigQuery. The team writes SQL exclusively. Which solution adds the least operational surface?

   • A.Cloud Composer DAG with BigQueryInsertJobOperator and an EmailOperator triggered on failure of the SQL task
   • B.BigQuery scheduled query plus a Cloud Monitoring alert with an email notification channel✓ Correct
   • C.Cloud Scheduler firing a Pub/Sub topic that wakes a Cloud Function which invokes the BigQuery Jobs API
   • D.Workflows execution chained behind Cloud Scheduler with a conditional notification step

   Why: The principle is that the simplest service meeting the requirement minimizes operational surface. A single SQL refresh on a fixed cadence is the native sweet spot of a scheduled query, and a Cloud Monitoring alert on the bigquery_dts failure metric covers the email notification. Cloud Composer is the most tempting distractor because email operators are a familiar Airflow idiom, but the hourly environment fee dwarfs the cost of the underlying refresh query.

4. Within Cloud KMS, which object directly contains a CryptoKey?

   • A.Key ring✓ Correct
   • B.A folder in the resource hierarchy
   • C.A service agent
   • D.A Cloud HSM partition

   Why: Cloud KMS organizes objects as project, key ring, key, key version. A key ring is the regional or multi-regional container that directly holds keys, and rotation creates versions inside each key. Folders sit in the resource hierarchy above projects and have nothing to do with key containment. A service agent is the principal that performs encryption on behalf of a data service; it is granted access to a key but does not contain keys. Cloud HSM partition is a protection level, not a hierarchy node.

5. Support dashboards render one-sentence summaries of 50,000 daily tickets. The first design called ML.GENERATE_TEXT inside a Looker tile, which produced an unbounded Vertex AI bill and high tile latency. The team needs sub-second tile loads with predictable daily cost while keeping the summarization quality. Which redesign fits the requirement?

   • A.Drop temperature from 0.9 to 0.2 inside the tile-bound ML.GENERATE_TEXT call
   • B.A persistent derived table in Looker that calls ML.GENERATE_TEXT on every datagroup refresh
   • C.A materialized view on the raw ticket text the tile reads instead of calling the model
   • D.A scheduled query that batches ML.GENERATE_TEXT against the prior-day partition into a destination table✓ Correct

   Why: Pre-computing the LLM output in a scheduled query against the prior-day partition writes a stable summary table that the tile reads in milliseconds, which caps Vertex AI cost at one batch per day. The constraint named sub-second tile loads, predictable daily cost, and the same summarization quality, which the pre-compute pattern satisfies. The Looker persistent derived table distractor is tempting since it also materializes results, but a datagroup refresh fires more often than once per day and still invokes the LLM during business hours.

GCP - Associate Data Practitioner Exam Details

Number of questions

60

Duration

120 minutes

Passing score

75%

Level

Associate

Delivered by

Google Cloud (GCP)

All figures should be confirmed on the official Google Cloud (GCP) page.

Frequently Asked Questions

Related certifications

• GCP - Professional Data Engineer
  Google Cloud (GCP)
• AWS SAA-C03
  Amazon Web Services (AWS)
• AZ-104
  Microsoft Azure
• SnowPro Core (COF-C03)
  Snowflake