Google Cloud Associate Data Practitioner - Fact Sheet¶

Exam Overview¶

Exam Name: Associate Data Practitioner Exam Code: None published on the exam page Duration: 2 hours Questions: 50-60 multiple choice and multiple select questions Passing Score: Not published on the exam page or guide Cost: $125 USD plus tax where applicable (exam page, read 2026-11-07) Delivery: Online proctored from home, or onsite at a Pearson VUE test center Languages: English, Japanese Valid For: Total validity is not stated on the exam page or the Renewal FAQ; the FAQ says a renewal adds 3 years to the inactive date (read 2026-11-07) Retakes: The FAQ says the waiting period between attempts varies by exam; none is published for this exam Prerequisites: None; 6+ months of hands-on experience with data on Google Cloud recommended Level: Associate

Registration is through Google Cloud's certification portal (the exam page links to cp.certmetrics.com). The renewal FAQ lists renewal terms for "other foundational and associate certifications": the renewal window opens 180 days before the inactive date, renewal adds 3 years, and the standard exam is the only option. Per the FAQ, you can also renew up to 30 days after the inactive date; after that you must retake the standard exam at full price. The Associate Cloud Engineer row was ambiguous in the fetched text, so read the FAQ on the date you plan to renew.

📖 Official exam page - format, fee, delivery, links to the guide and sample questions 📖 Exam guide v1.0 (PDF) - the full section and objective list 📖 Renewal FAQs - renewal window and extension 📖 Training path - Google Cloud Skills Boost path for the exam

Sections and Weights¶

Target Audience¶

• Analysts and junior data engineers who work with data on Google Cloud
• People who need to ingest, transform, analyze, orchestrate and secure data, not design large systems
• A step before the Professional Data Engineer certification

Section 1: Data Preparation and Ingestion (~30%)¶

📖 Storage Transfer Service - online transfers, optimized for more than 1 TiB 📖 Transfer Appliance - offline device when upload takes over a week 📖 BigQuery Data Transfer Service - scheduled loads into BigQuery 📖 Cloud Storage locations - region, dual-region, multi-region

Key Facts: - ELT loads raw data first and transforms in the warehouse; ETL transforms first; ETLT is a hybrid - Storage Transfer Service: S3, Azure Blob, on-premises, HDFS, URLs into Cloud Storage; Transfer Appliance when upload takes more than a week - BigQuery batch loads accept Avro, CSV, JSON, ORC and Parquet; Avro and Parquet carry their schema - Region is cheapest; dual-region gives cross-region redundancy; multi-region spans a large geography; turbo replication targets 15 minutes - Cloud Data Fusion is visual; Dataflow is code or templates on Apache Beam

Section 2: Data Analysis and Presentation (~27%)¶

📖 Introduction to BigQuery ML - model types and remote models 📖 ML.EVALUATE - evaluation metrics 📖 Generative AI in BigQuery - remote models and AI functions

Key Facts: - BigQuery ML trains models with SQL: CREATE MODEL (MODEL_TYPE required), ML.EVALUATE, ML.PREDICT - Classification: logistic regression or boosted trees; regression: linear; clustering: K-means; recommendation: matrix factorization; time series: ARIMA_PLUS - ML.EVALUATE classification metrics: precision, recall, accuracy, F1, log loss, ROC AUC; default threshold 0.5 - Remote models call hosted models such as Gemini; AI functions run over columns - Looker uses LookML for governed shared metrics; Looker Studio is self-service reporting

Section 3: Data Pipeline Orchestration (~18%)¶

📖 Scheduling queries - scheduled query rules 📖 Cloud Composer - managed Airflow 📖 Workflows - serverless orchestration 📖 Eventarc - event routing 📖 Pub/Sub BigQuery subscriptions - direct writes to BigQuery

Key Facts: - Scheduled queries: minimum interval 5 minutes; @run_time and @run_date; avoid exactly on the hour - Composer = Airflow DAGs in Python; Workflows = YAML or JSON steps calling services; pick the simplest tool - Dataform = SQLX plus ref() dependencies and assertions - BigQuery subscription = at-least-once, no processing; use Dataflow for windowing or complex transforms - Eventarc destinations: Cloud Run, Workflows, Cloud Run functions

Section 4: Data Management (~25%)¶

📖 BigQuery IAM roles - predefined roles 📖 Cloud Storage classes - minimum durations 📖 Object Lifecycle Management - rules and actions 📖 Cloud SQL high availability - standby and failover 📖 Cloud KMS - CMEK and keys

Key Facts: - Least privilege: Data Viewer to read, Job User to run queries, avoid basic roles - Uniform bucket-level access is recommended; fine-grained ACLs raise exposure risk - Classes: Standard none, Nearline 30 days, Coldline 90 days, Archive 365 days; Autoclass automates - Lifecycle actions: Delete, SetStorageClass, AbortIncompleteMultipartUpload; changes take up to 24 hours - Cloud SQL HA: standby in another zone, synchronous writes, about 60 seconds failover, standby serves no reads - GMEK default; CMEK in Cloud KMS; CSEK supplied by you and not a KMS feature - Analytics Hub shares datasets without copying, as read-only linked datasets

Exam Tips¶

1. Pick the simplest tool that meets the requirement: a scheduled query before Composer, a BigQuery subscription before Dataflow.
2. Match storage class to access frequency: monthly, quarterly, yearly.
3. Least privilege wording points to predefined roles at the narrowest scope.
4. HA, replicas and backups are different tools: failover, read scaling, human error.
5. Time management: about 2 minutes per question across 50-60 questions.

Quick Reference¶

-- BigQuery ML (not run: needs a BigQuery project)
CREATE OR REPLACE MODEL `proj.ds.m` OPTIONS (model_type = 'LOGISTIC_REG', input_label_cols = ['label'])
AS SELECT * FROM `proj.ds.train`;
SELECT * FROM ML.EVALUATE(MODEL `proj.ds.m`, (SELECT * FROM `proj.ds.test`));
SELECT * FROM ML.PREDICT(MODEL `proj.ds.m`, (SELECT * FROM `proj.ds.new_rows`));