CertifyCloudxAce every cloud certification.

PCDBE sample questions with answers

10 free practice questions for the Professional Cloud Database Engineer exam. Try each one, then open the answer to see why the right option wins and every other option loses.

Question 1Design innovative, scalable, and highly available cloud database solutions

Nordhavn Insurance must not store or process policy data outside the European Union and enforces the resource locations policy accordingly. A product team needs a relational database with zero data loss if a region is lost and strongly consistent global reads, and has proposed a Spanner instance in a North American multi-region configuration. What should you do?

  1. A.

    Deploy a regional Spanner instance in europe-west4 and copy backups to a United States bucket.

  2. B.

    Create the Spanner instance in an EU multi-region configuration, with its Cloud KMS keys in that configuration's EU regions.

  3. C.

    Request an exception relaxing the resource locations constraint, then deploy the North American multi-region instance as originally designed.

  4. D.

    Deploy the North American instance but encrypt policy data in the application with an EU key.

Show answer

Answer: B

The organization policy and the residency rule are hard boundaries, so the design must use a multi-region configuration whose replicas all sit in allowed EU regions.

  • A. A regional instance cannot survive the loss of its region without data loss, and copying backups to the United States violates residency.
  • B. An EU multi-region configuration with keys in the replica regions satisfies residency and the organization policy while still delivering zero data loss and strong consistency.
  • C. Relaxing the guardrail to fit a design breaks the supervisory rule the policy implements and would fail audit, regardless of technical feasibility.
  • D. Storing ciphertext outside the EU is still storing the data outside the EU, so application-level encryption does not satisfy a residency requirement.
Question 2Design innovative, scalable, and highly available cloud database solutions

Arlen Wind Power collects 1.5 million sensor readings per second from turbines. Control-room screens fetch the last hour for one turbine in under 20 milliseconds, while analysts run ad hoc SQL over three years of history against weather data already in the warehouse. Each workload should be served by the technology suited to it. Which two components should you include in the design? (Choose two.)

Choose 2.

  1. A.

    Bigtable as the serving store, keyed by turbine identifier plus reversed timestamp for contiguous scans.

  2. B.

    Cloud SQL for PostgreSQL as one store for both workloads, with a composite index.

  3. C.

    BigQuery as the analytical store, loaded continuously and partitioned by date for multi-year joins.

  4. D.

    Firestore as the serving store, with one document per reading and collection group queries.

  5. E.

    Memorystore for Redis as the analytical store, holding three years in sorted sets.

Show answer

Answer: A, C

Polyglot persistence pairs Bigtable for high-throughput low-latency time-series serving with BigQuery for large-scale ad hoc SQL over history.

  • A. Bigtable sustains the ingest rate and serves narrow range scans in milliseconds when the row key leads with the turbine and reverses the timestamp.
  • B. A single-primary relational instance cannot absorb 1.5 million writes per second, and multi-year scans on it would be slow and expensive.
  • C. BigQuery handles multi-year ad hoc SQL and joins to the existing weather tables, and date partitioning limits the data each query scans.
  • D. Firestore is not designed for this sustained write rate, and one document per reading would make both cost and query latency unacceptable.
  • E. Memorystore keeps the entire dataset in memory, so three years of readings is cost-prohibitive and it offers no ad hoc SQL for analysts.
Question 3Design innovative, scalable, and highly available cloud database solutions

Birchwood Books runs Cloud SQL for MySQL 8.0 on a current maintenance version. It wants to store product embeddings and run approximate nearest-neighbour searches in the same database as its catalogue tables, rather than adopt a new database product. What should you do?

  1. A.

    Install the pgvector extension on the MySQL instance, and then add a vector column to the catalogue table.

  2. B.

    Enable the cloudsql_vector database flag, store the embeddings in VECTOR columns, and create a vector index for ANN searches.

  3. C.

    Upgrade the instance to a newer MySQL major version, because 8.0 cannot store vector embeddings.

  4. D.

    Keep the embeddings in Memorystore for Valkey, because Cloud SQL for MySQL cannot store vectors.

Show answer

Answer: B

Cloud SQL for MySQL 8.0 and 8.4 support vector embeddings when the cloudsql_vector flag is enabled, and ANN search requires a vector index.

  • A. pgvector is a PostgreSQL extension and cannot be installed on MySQL.
  • B. On MySQL 8.0 and 8.4, cloudsql_vector enables vector storage, and a vector index is required for ANN searches.
  • C. MySQL 8.0 supports vector embeddings in Cloud SQL once the cloudsql_vector flag is enabled.
  • D. Cloud SQL for MySQL can store vectors in ACID tables alongside the other data.
Question 4Design innovative, scalable, and highly available cloud database solutions

Moorfield Tickets serves live seat inventory from Bigtable. Analysts now need every change to that table available in BigQuery within minutes, continuously, so that they can study booking patterns. The team does not want to write or maintain custom pipeline code. What should you do?

  1. A.

    Create a Datastream stream with the Bigtable table as its source and BigQuery as its destination.

  2. B.

    Export the Bigtable table to Cloud Storage every night, and load each export into BigQuery.

  3. C.

    Create a BigQuery external table over the Bigtable table, and treat it as the continuous change history.

  4. D.

    Enable a change stream on the Bigtable table, and run the Google-provided Bigtable change streams to BigQuery Dataflow template.

Show answer

Answer: D

Bigtable change streams provide change data capture, and the Google-provided Dataflow template streams those changes into BigQuery without custom code.

  • A. Datastream does not list Bigtable among its sources; Bigtable provides change streams for CDC.
  • B. Nightly exports are neither continuous nor within minutes.
  • C. An external table shows the current contents; it does not record the sequence of changes over time.
  • D. Change streams capture every data change, and the Google-provided template writes them to BigQuery without custom code.
Question 5Design innovative, scalable, and highly available cloud database solutions

Orlick Telecom plans to move a 40-node HBase cluster of call-detail records to Bigtable and switch its Java services to the Bigtable HBase client. A dependency review finds that the services rely on HBase coprocessors for server-side aggregation and on per-cell visibility labels that hide some fields from contractors. What should you conclude?

  1. A.

    The coprocessors carry over through the HBase client, and only the per-cell visibility labels need to be replaced.

  2. B.

    Both features carry over unchanged, because the Bigtable HBase client for Java implements the full HBase 2.x API surface.

  3. C.

    The coprocessor JAR files can be deployed to the Bigtable instance so that the aggregation keeps running on its nodes.

  4. D.

    Both features need redesign, such as moving the aggregation and field filtering into the services, because Bigtable supports neither.

Show answer

Answer: D

Moving HBase code to Bigtable keeps most of the API, but coprocessors and per-cell visibility are unsupported, so dependencies on them must be redesigned before the move.

  • A. Coprocessors are on Bigtable's list of unsupported HBase features, so the aggregation can't move with the client either.
  • B. The HBase client is compatible with the HBase API but lists unsupported features, including coprocessors and cell visibility.
  • C. Bigtable runs no customer code on its nodes; coprocessor classes can't be created or loaded.
  • D. Bigtable documents that coprocessors aren't supported and that you can't set the visibility of individual cells, so both dependencies need another design.
Question 6Design innovative, scalable, and highly available cloud database solutions

Corvane Payments runs its transaction ledger in a GoogleSQL-dialect database on a Spanner Enterprise edition instance. The fraud team wants to find rings of accounts linked through shared devices and cards up to five hops away, with results consistent with the latest committed transactions and without copying data to another system. What should you do?

  1. A.

    Move the instance down to the Standard edition, which also includes Spanner Graph, and then query the graph with GQL.

  2. B.

    Define a property graph over the existing Spanner tables, and query it with Spanner Graph by using GQL.

  3. C.

    Replicate the tables into Bigtable, and walk the links from the application with row-key lookups.

  4. D.

    Export the tables to BigQuery every night, and find the linked accounts with recursive SQL queries.

Show answer

Answer: B

Spanner Graph lets you define a property graph over existing Spanner tables and run multi-hop GQL queries on current data, which suits relationship-heavy fraud detection.

  • A. Spanner Graph is available only in the Enterprise and Enterprise Plus editions, not in Standard.
  • B. Spanner Graph maps existing tables to a property graph without data migration and queries it with GQL on the live, consistent data.
  • C. This copies the data and pushes a five-hop traversal into application code.
  • D. A nightly copy is up to a day stale and is the second system the requirement excludes.
Question 7Design innovative, scalable, and highly available cloud database solutions

Rowdon Payments copies card transactions from Cloud SQL into BigQuery for fraud analytics. Its PCI DSS scope review requires that the analytics copy never contain real card numbers, while analysts must still count and join transactions by the same card across tables. The team already uses Sensitive Data Protection (formerly Cloud DLP) for discovery. What should you do?

  1. A.

    De-identify the card numbers with deterministic encryption, such as AES-SIV, before the data lands in BigQuery.

  2. B.

    Run Sensitive Data Protection discovery on the BigQuery dataset, and alert whenever it finds card numbers.

  3. C.

    Replace every card number with a random value during the copy, so that no token can be linked back to a card.

  4. D.

    Encrypt the BigQuery dataset with a customer-managed key, and let analysts query the real card numbers.

Show answer

Answer: A

Sensitive Data Protection's deterministic pseudonymization tokenizes card numbers consistently, so analytics keep referential integrity without holding real values.

  • A. Deterministic pseudonymization replaces each card number with the same token every time, preserving joins without exposing real values.
  • B. Discovery finds sensitive data but does not remove it from the analytics copy.
  • C. Random replacement breaks referential integrity, so analysts could no longer count or join by card.
  • D. CMEK protects storage at rest; analysts would still see real card numbers, which keeps the copy in scope.
Question 8Design innovative, scalable, and highly available cloud database solutions

Otterburn Retail keeps 80 million product listings from marketplace sellers in BigQuery. Each night it must find listings whose descriptions are semantically near-duplicates of existing ones and write the pairs to a table for review. No online application reads the results, and the team wants to avoid moving the data. What should you do?

  1. A.

    Copy the listings to Memorystore for Valkey every night, and run vector search there from a batch job.

  2. B.

    Load the listings into Cloud SQL for PostgreSQL with pgvector, and compare every pair of rows in SQL.

  3. C.

    Export the listings to Firestore, and run a nearest-neighbour query for each of the 80 million documents.

  4. D.

    Generate embeddings with AI.GENERATEEMBEDDING in BigQuery, and find near neighbours with VECTORSEARCH.

Show answer

Answer: D

For batch semantic similarity over data already in BigQuery, generate embeddings with AI.GENERATEEMBEDDING and search them with VECTORSEARCH, optionally with a vector index.

  • A. It moves 80 million listings into memory for a batch task that needs no low latency.
  • B. It moves the data into an operational database and compares all pairs, which does not scale.
  • C. It moves the data and turns one batch job into 80 million separate queries.
  • D. BigQuery can generate embeddings and run vector search in SQL where the listings already are.
Question 9Design innovative, scalable, and highly available cloud database solutions

Radley Studios enforces the Cloud SQL public IP restriction at the organization level. Security has approved an exception for one lab folder, where a vendor's hosted tool can connect only over public IP with authorized networks. Every other folder, including any created later, must stay protected. What should you do?

  1. A.

    Remove the organization-level policy, and enforce the constraint separately on each of the other folders.

  2. B.

    Place the lab folder inside a VPC Service Controls perimeter, which exempts it from organization policies.

  3. C.

    Grant the lab team the Organization Policy Administrator role so that it can turn the constraint off as needed.

  4. D.

    Set a policy on the lab folder that does not enforce the constraint, overriding the inherited organization policy.

Show answer

Answer: D

Organization policies are inherited, but a policy set on a folder supersedes the parent's, so a scoped exception belongs on the lab folder while the organization-level policy keeps protecting everything else.

  • A. Folders created later would start unprotected unless someone remembered to add the policy.
  • B. Service perimeters do not exempt resources from organization policy constraints.
  • C. It hands the power to change guardrails organization-wide to the team that wants the exception.
  • D. A policy set on a resource supersedes its parent's, so only the lab folder and its projects are exempted.
Question 10Design innovative, scalable, and highly available cloud database solutions

Linford Travel is building a new Go backend on Cloud Run that stores itineraries as documents. An engineer proposes Firestore in Datastore mode because the service runs only on the server. The team has no existing Datastore code, and product managers expect real-time updates in a future web client. What should you do?

  1. A.

    Create the Firestore database without Datastore compatibility, and use the Firestore server client library for Go.

  2. B.

    Create the database in Datastore mode, because Datastore mode is the recommended choice for server-side workloads.

  3. C.

    Use Cloud SQL for MySQL with a JSON column, because document databases are unsuited to server-side code.

  4. D.

    Create the database in Datastore mode, and add Firestore real-time listeners when the web client ships.

Show answer

Answer: A

Datastore compatibility is only for apps that depend on the Datastore APIs; a new server-side service should use Firestore directly to keep real-time and richer query features.

  • A. Google recommends Datastore compatibility only for apps that depend on the Datastore APIs; Firestore itself serves server-side backends and keeps real-time features.
  • B. Current guidance is the reverse: use Datastore compatibility only for legacy Datastore API dependencies.
  • C. Firestore is designed for server-side backends as well as mobile and web apps, so the premise is false.
  • D. Datastore compatibility disables Firestore real-time capabilities, so the listeners would not be available.

Keep going with 490 more PCDBE questions

Free papers every day, in the real exam formats, with progress by exam domain. Unlock every paper and timed mock exam when you are ready.

Practise freeSee all PCDBE papers
PCDBE sample questions with answers (10 free) · CertifyCloudx