Read this the night before or on exam morning. It assumes you’ve studied the notes. It’s here to refresh, not to teach.
Exam day
- 50 to 60 multiple choice and multiple select questions in 2 hours. Registration fee is $125 plus tax where applicable. English and Japanese. No prerequisites; Google recommends 6+ months of hands-on experience with data on Google Cloud.
- Google doesn’t publish a passing score for this exam. You get a pass or fail result. Roughly 2 minutes per question; flag and move on.
- Section weights (official guide): data preparation and ingestion about 30%, data analysis and presentation about 27%, data management about 25%, data pipeline orchestration about 18%.
How to read a question
Read the last sentence first, then find the constraint words.
| The question says | Pick |
|---|---|
| Least operational overhead, fully managed, serverless | Dataflow, BigQuery, Dataform, Workflows, scheduled queries over clusters or custom code |
| SQL team, data already in BigQuery | BigQuery SQL, Dataform, BigQuery ML |
| No code, visual | Cloud Data Fusion, Looker Studio, AutoML |
| Real time, streaming | Pub/Sub, Dataflow, BigQuery subscription |
| Least privilege | Predefined role at the lowest level; never Owner or Editor |
| Cost-effective, rarely accessed | Colder storage class, lifecycle rule, partition expiration |
| Without copying the data | Analytics Hub, authorized views, external tables |
Elimination: answers that move data you could query in place, add a cluster to manage, or grant basic roles are usually wrong.
Tool-choice tables
| Moving data | Tool |
|---|---|
| Files, online, S3 or on-prem to Cloud Storage | Storage Transfer Service |
| Petabytes, slow network | Transfer Appliance |
| SaaS or ads platforms into BigQuery on a schedule | BigQuery Data Transfer Service |
| Database to Cloud SQL or AlloyDB | Database Migration Service |
| Continuous database change capture | Datastream |
| Messages straight into a table | Pub/Sub BigQuery subscription |
| Transforming data | Tool |
|---|---|
| Streaming or large Beam pipelines, serverless | Dataflow |
| Existing Spark/Hadoop jobs | Dataproc (Managed Service for Apache Spark) |
| Visual, low-code | Cloud Data Fusion |
| SQL ELT inside BigQuery with dependencies | Dataform |
| One recurring query | Scheduled query |
| Orchestrating | Tool |
|---|---|
| Complex DAG across services | Cloud Composer (Airflow) |
| Few API calls in order, serverless | Workflows |
| Only Dataproc jobs | Dataproc Workflow Template |
| Cron trigger for HTTP, Pub/Sub | Cloud Scheduler |
| React to a file or event | Eventarc to Cloud Run, Cloud Run functions, Workflows |
Storage and formats
| Need | Service |
|---|---|
| Files, data lake | Cloud Storage |
| SQL analytics | BigQuery |
| Relational app, regional | Cloud SQL |
| Relational, global scale | Spanner |
| Huge key-based, low latency | Bigtable |
| Documents for apps | Firestore |
- Avro and Parquet load faster and keep the schema; Parquet is columnar. CSV and JSON have no types; BigQuery needs newline-delimited JSON.
- Location types: zonal, regional, dual-region, multi-region. Can’t be changed after bucket creation.
- BigQuery load jobs are free; streaming is billed.
BigQuery cost and ML
SELECT *scans all columns;LIMITdoesn’t cut bytes. Partition on a date, cluster on filter columns, use a dry run, cached results are free for about 24 hours.- Time travel 7 days by default; long-term storage rate after 90 days unmodified.
- ML:
CREATE MODEL,ML.EVALUATE,ML.PREDICT. Classification: logistic regression or boosted trees. Regression: linear. Clustering: k-means. Forecasting: ARIMA_PLUS. LLMs: remote model over a connection. - Looker Studio is free and quick; Looker has the LookML governed model.
Storage classes
| Class | Minimum duration |
|---|---|
| Standard | none |
| Nearline | 30 days |
| Coldline | 90 days |
| Archive | 365 days |
All have millisecond access and the same durability; colder means cheaper storage and costlier retrieval. Autoclass moves objects automatically. Lifecycle rules: Delete or SetStorageClass by age.
Access and security
- Roles: basic (avoid), predefined (default), custom. BigQuery Job User runs queries but doesn’t read data; add Data Viewer.
- Cloud Storage: uniform bucket-level access (IAM only), signed URLs for temporary access, public access prevention.
- Analytics Hub: exchange, listing, linked dataset, no copies.
- Encryption: all data encrypted at rest and in transit by default. GMEK default, CMEK in Cloud KMS for control and revocation, CSEK only for Cloud Storage and Compute Engine disks (lose the key, lose the data).
- HA (standby in another zone) is not a backup. Backups and point-in-time recovery fix deletion; read replicas and cross-region replicas help regional DR.