Prepare and test your skills

Prepare and test your skills

Free sample question with answers and detailed explanation below. The Professional Cloud Database Engineer practice bank has an additional 769 practice questions.Try 10 free realistic questions: instant score, no credit card, and no account required.
A healthcare enterprise is building a Retrieval-Augmented Generation (RAG) clinical assistant using AlloyDB for PostgreSQL with the pgvector-compatible vector extension to store patient record embeddings alongside structured metadata. During peak clinical hours, high-concurrency filtered similarity searches overload the primary instance, and security policies mandate strict data privacy during retrieval.
Your solution must meet the following requirements:
Which configuration should you implement?
The Gen AI evaluation service assesses the quality of LLM outputs against metrics such as factual accuracy, coherence, relevance, and grounding. Model Armor acts as an AI security gateway that screens inbound prompts and outbound responses for prompt injection, jailbreak attempts, and harmful content, while Sensitive Data Protection (SDP) discovers, classifies, and de-identifies personally identifiable information (PII) in datasets and prompts.
ScaNN index scans or lock contention.The primary goal were evaluating RAG output quality, sanitizing user prompts against injection attacks, and redacting PII from raw documents prior to embedding ingestion, rather than scaling and securing the underlying vector database tier.
This configuration combines AlloyDB for PostgreSQL vector acceleration features, horizontal read scaling, database observability, and perimeter-level security controls to support production Retrieval-Augmented Generation (RAG) workloads. By pairing Approximate Nearest Neighbor (ANN) indexing with dedicated read infrastructure and zero-trust network and identity controls, the database tier handles heavy similarity searches reliably without exposing sensitive embeddings.
ScaNN index (Scalable Nearest Neighbors) alongside the columnar engine accelerates vector distance calculations and metadata-filtered scans through adaptive filtering and inline filtering.ScaNN index provides up to 10x faster filtered vector search queries and index builds compared to standard PostgreSQL HNSW indexes.SQLcommenter traces slow vector queries directly back to application code.Unlike vertical scaling or passive high-availability standbys, read pool instances scale read throughput horizontally, while the ScaNN index and columnar engine directly optimize hybrid SQL-plus-vector workloads. Simultaneously, Auth Proxy and VPC Service Controls enforce database-layer governance that application-layer filters alone cannot provide.
Cloud SQL for PostgreSQL Enterprise Plus provides a local SSD-backed data cache that stores frequently accessed data pages to accelerate read-heavy queries, including pgvector similarity searches. Cloud Monitoring exposes specialized cache metrics—such as database/postgresql/data_cache/hit_ratio, hit_count, and miss_count—to evaluate cache efficiency, while authorized networks allowlist external IPv4 CIDR ranges for public-IP database connections.
IVFFlat index sacrifices AlloyDB read pool instances, the ScaNN index, and the columnar engine, which are specifically engineered for high-concurrency filtered vector scans.database/postgresql/data_cache/hit_ratio helps tune single-node Cloud SQL memory and SSD cache sizing, but it does not scale read compute across a cluster of nodes.IVFFlat indexes generally require careful list tuning and offer lower recall-to-latency efficiency than ScaNN on large, dynamic datasets.The workload already ran on Cloud SQL for PostgreSQL Enterprise Plus, required single-instance local SSD read acceleration monitored via hit_ratio metrics, and did not require AlloyDB's ScaNN engine, IAM connection proxies, or VPC Service Controls perimeters.
An AlloyDB high-availability (HA) cluster maintains an active node and a passive standby node across zones to guarantee rapid automated failover during infrastructure outages. VPC firewall rules operate at the network transport layer (Layer 3 and Layer 4) to permit or deny IP traffic between virtual network endpoints, while static credentials authenticate database sessions using stored usernames and passwords.
HNSW alone misses the substantial filtered-search performance gains of ScaNN and the columnar engine.HNSW indexes without ScaNN inline filtering or columnar acceleration incur higher latency when combining metadata predicates with vector similarity.The primary requirement were strictly zonal fault tolerance with minimal connection-level changes in an isolated single-project environment that did not experience primary-node read bottlenecks or require IAM connection governance.
A common misconception is treating an HA standby instance as an active read replica. In managed PostgreSQL services on Google Cloud, horizontal read scaling requires dedicated read pool instances (in AlloyDB) or read replicas (in Cloud SQL), rather than relying on the HA standby.
Answer them yourself and get your score the moment you finish. No account needed.
Full access includes
Keep the momentum going with these hand-picked practice scenarios
Last updated Validated