Which Google Cloud Database Should You Use? A Study Guide for the Database Engineer Exam

On the Google Cloud Professional Cloud Database Engineer exam, most questions come down to choosing the right managed database for a described workload, so the fastest way to pass is to learn each database by its disqualifiers. Cloud SQL and AlloyDB are your relational defaults, Spanner is for global scale with strong consistency, Bigtable is for massive high-throughput non-relational workloads, Firestore is for document data and mobile sync, and Memorystore is a cache, not a system of record. Know when each one is the wrong answer and the scenario questions almost solve themselves.

The exam rarely asks what a database is. It describes scale, consistency needs, latency budget, data shape, and access pattern, then offers several databases as options. This guide is a decision map for that exact situation.

The six you must be able to choose between

The decision questions the exam is really asking

Is the data relational? If yes, you are choosing between Cloud SQL, AlloyDB, and Spanner. If no, you are in Bigtable or Firestore territory. This first cut eliminates half the options on most questions.

Does it need to scale writes horizontally or span regions with strong consistency? That is Spanner's signature, and it is usually the differentiator when a question mentions global users and strong consistency in the same breath. If the scenario is single region and modest scale, Spanner is the expensive wrong answer.

Is throughput the headline? When a question describes millions of reads and writes per second, time series, or sensor data, Bigtable is almost always the intended answer, and any mention of SQL joins or cross-row transactions rules it out.

Is it a mobile or web app with flexible documents? Firestore, especially when offline sync or real-time listeners appear in the requirements.

Is it AI or vector search? This is the newest wrinkle. The current exam guide added vector databases tied to generative AI and large language model use cases, and AlloyDB is the featured option for vector search on relational data. Older study material predates this entirely, so if your prep never mentions vectors, it has a gap in the section that carries the most weight.

Migration matters as much as selection

Choosing the database is half the job. The exam devotes roughly 23 percent to migrating data solutions, so you also need to reason about Database Migration Service, homogeneous versus heterogeneous moves, minimizing downtime at cutover, and validating that the target matches the source. A common question pattern gives you a source database and a target and asks for the migration approach with the least downtime and risk. Knowing the difference between a lift-and-shift homogeneous move and a heterogeneous conversion, and how you would verify data integrity afterward, is what these questions test.

When workloads span several of these databases

Real systems rarely use one database, and the exam's second-largest section, managing a solution that spans multiple database technologies, reflects that. Once your data lives in a mix of Spanner, Bigtable, and Firestore, operational questions get harder: consistent monitoring, access control, and letting people actually find information that is scattered across engines. In production that last problem is often solved by putting a single search layer over data spread across every system, and the exam wants you thinking at that fleet level rather than tuning one database in isolation.

A worked example, the way the exam phrases it

Consider a question like this: a global gaming company needs a database for player profiles and inventory, with strongly consistent reads and writes across regions, sustained high transaction volume, and horizontal scaling as the game grows. Which database fits? Walk the decision map. The data is relational, so you are choosing among Cloud SQL, AlloyDB, and Spanner. The requirement names global scope, strong consistency, and horizontal write scaling in one sentence, which is Spanner's exact signature. Cloud SQL cannot scale writes past a single primary, and AlloyDB is regional, so both are the plausible-but-wrong distractors. The answer is Spanner.

Now change one detail. Same company, but the workload is billions of time-series events from game telemetry, queried by key with no need for joins. The data shape flipped to non-relational and high throughput, which rules out every relational option and points straight at Bigtable. The exam builds most of its questions this way: a scenario with one or two decisive details buried among neutral ones. Training yourself to spot the decisive detail is the whole skill.

Turn this map into practice questions

Reading a decision table once does not build recall. The skill the exam tests is retrieving the right choice under a scenario, and the only way to train that is repetition against realistic prompts. Write yourself workload descriptions, or better, turn your own study notes and the current exam guide into questions so the choices come back automatically on exam day. You can generate Professional Cloud Database Engineer practice questions from your own notes that mirror this select-the-database format across all four sections.

Master the disqualifiers, keep the migration patterns fresh, and remember that Google publishes no passing score, so the goal is broad competence rather than a magic percentage. Do that and the database-selection questions, which are the heart of this exam, become the part you look forward to.