Storage

GCP Database Decision Tree: How to Choose for PCA Exam

February 5, 2026 13 min read Last reviewed: October 6, 2026 By GCP Architect Team

The Short Answer

Choose the database from the workload, not the brand. Cloud SQL for single-region relational OLTP, Cloud Spanner when you need relational data with global strong consistency, Cloud Bigtable for massive write-throughput NoSQL telemetry, Firestore for real-time mobile and web apps, and BigQuery for analytical queries over large datasets. The flowchart below walks the same decision the exam expects.

If there is one topic that defines the PCA exam, it's Database Selection. Google doesn't just want to know if you know the services—they want to know if you know when to use them. For a broader overview of exam success, see our top 10 PCA tips.

GCP Database Decision Tree Flowchart

Follow these questions to find your match:

1. Is your data structured or unstructured?

  • Unstructured (Files/Videos): Cloud Storage.
  • Structured/Semi-structured: Proceed to next question.

2. Is it for Analytics (OLAP) or Transactions (OLTP)?

  • Analytics (Big Data/SQL): BigQuery.
  • Transactions/Operational: Proceed to next question.

3. Do you need relational (SQL) or NoSQL?

  • Relational:
    • Normal scale/Single region: Cloud SQL.
    • Massive scale/Global/99.999% SLA: Cloud Spanner (Highly recommended for cases like Mountkirk Games).
  • NoSQL:
    • Document/Mobile/Web sync: Firestore.
    • High-throughput/Low-latency/Analytical-lite: Bigtable (Essential for TerramEarth IoT data).

Deep Dive: The SQL Choices

Service Key Strengths Max Scale
Cloud SQL Standard MySQL/Postgres, ease of use Up to 64TB
Cloud Spanner Horizontal scale, Strong Consistency Petabytes+

Where the Decision Tree Breaks Down

A decision tree gets you to a shortlist. It does not get you to an answer, and the exam questions that catch people are the ones where two services on the shortlist both fit the stated requirement.

The first place it breaks down is cost at small scale. Cloud SQL and Spanner both offer relational semantics with strong consistency, so the tree can point at either one. Spanner's minimum footprint makes it expensive for a workload that a single Cloud SQL instance would handle comfortably, and a question that mentions a small regional deployment is testing whether you noticed that.

The second place is write volume. Firestore and Bigtable both handle unstructured data at scale, but they diverge sharply once writes become heavy and time-ordered. Bigtable is built for that pattern. Firestore is not, and picking it because the data "looks unstructured" is the most common wrong answer in this domain.

The third is multi-region. Cloud SQL offers high availability within a region. Spanner offers consistency across regions. If the requirement says global and strongly consistent, the tree has one branch, and every other option is a distractor.

Two Habits That Make These Questions Easier

Two habits make these questions easier. The first is reading for the constraint that eliminates options rather than the one that sounds most impressive. Questions are usually built around a single hard requirement, and the remaining detail is atmosphere.

The second is knowing which features are genuinely unique. Spanner's external consistency, Bigtable's row-key design, and BigQuery's separation of storage and compute are the three that show up most in scenario questions. If you can state what each one does that no other service on the list does, most of these questions resolve without a flowchart.

Which Database for Each Case Study

The PCA case studies make this concrete, and the database question usually arrives attached to a specific workload rather than as an abstract choice.

Mountkirk Games describes a gaming platform with time-series telemetry at high write volume. That is Bigtable, and the row-key design is the reason. The question is testing whether you would design the key around the query pattern rather than accepting a default.

TerramEarth describes sensor data arriving from heavy equipment in the field, written continuously and analysed in bulk later. The answer is usually Bigtable for the hot path and BigQuery for the analytical path, with the split itself being the point. Candidates who name only one service miss that the scenario has two workloads.

EHR Healthcare describes regulated patient records with strong consistency requirements. That is Cloud SQL at small scale and Spanner once the compliance requirement includes multi-region availability, because neither Bigtable nor Firestore offers the transactional guarantees the scenario implies.

The habit worth building is reading the case study for the access pattern before the data type, because the data type only narrows the list while the access pattern is what actually picks the answer. The data type narrows the list. The access pattern picks the answer.

The One Thing to Remember

If you remember one thing from this page, make it the ordering: the access pattern decides, the data type only narrows, and cost eliminates whatever survives both.

Frequently Asked Questions (FAQ)

What is the maximum size for Cloud SQL?

Cloud SQL currently supports up to 64TB of storage, making it suitable for most standard relational workloads.

When is Cloud Bigtable the right choice?

Bigtable is ideal for low-latency, high-throughput NoSQL workloads like IoT telemetry, financial trade history, or ad-tech data.

Does Google Cloud provide a graph database?

Yes, Google Cloud provides Managed Service for Apache Age (PostgreSQL extension) and integrates with partners like Neo4j for dedicated graph workloads.

Exam Cheat Sheet:
  • Firestore = Mobile apps / Catalogs.
  • Bigtable = IoT streams / Time series / Ads.
  • Spanner = Global Finance / Inventory.
  • BigQuery = Business Intelligence / Logs.

Built for Architect Success

Take a practice exam to see these decision trees in action with real-world scenarios.

Start Practice Exam