Deploying Data Solutions
Google ACE ACE · Planning and implementing a cloud solution
Deploying data products
The exam guide's data section covers deploying the managed data services — Cloud SQL, Firestore, BigQuery, Spanner, Pub/Sub, Dataflow, Cloud Storage, and AlloyDB — and loading data into them. For the ACE exam, "deploying" a data product means creating the instance, dataset, or topic with the right configuration: region, machine size or edition, high-availability settings, backup settings, and access controls. You are not expected to be a database administrator, but you are expected to know what each service is for, how to create it, and how data gets in.
Relational databases: Cloud SQL and AlloyDB
Cloud SQL is the fully managed relational database for MySQL, PostgreSQL, and SQL Server. Creating an instance means choosing the database engine and version, the region and zone, the machine type and storage size, and the availability setting. The key availability choice is regional (high availability) versus zonal: a regional instance maintains a standby in another zone in the same region with automatic failover, while a zonal instance is cheaper but goes down with its zone. Automated backups and point-in-time recovery are configurable at creation and should be enabled for anything important. Cloud SQL instances can be reached over a public IP (with authorized networks restricting access) or a private IP inside the VPC — private IP is the recommended, more secure option.
AlloyDB is Google's PostgreSQL-compatible database for demanding workloads: higher performance and availability than Cloud SQL for PostgreSQL, with features such as columnar storage for analytics. For the ACE exam, the important distinction is that AlloyDB is PostgreSQL-compatible (not MySQL or SQL Server) and is the answer when a PostgreSQL workload needs more performance than Cloud SQL provides. Both services support read replicas for scaling read traffic.
Firestore, Spanner, and BigQuery
Firestore is deployed as a database in either Native mode (the document database with real-time sync) or Datastore mode (for compatibility with the older Datastore). The choice is made at creation and cannot be changed afterward. Firestore databases are regional or multi-regional; multi-regional gives higher availability for globally accessed data. Spanner is deployed as an instance with a configuration (regional or multi-regional) and a compute capacity measured in processing units or nodes; its defining features are global distribution and strong consistency. BigQuery is deployed as datasets within a project, in a chosen location (regional, multi-regional, or US/EU) — and a dataset's location cannot be changed after creation, which is a classic exam trap. BigQuery is serverless: there are no instances to size, and you pay for the data stored and the data processed by queries.
Messaging and processing: Pub/Sub and Dataflow
Pub/Sub is deployed as topics and subscriptions. A topic is the named channel publishers send messages to; a subscription is how a subscriber receives them — pull (the subscriber requests messages), push (Pub/Sub delivers them to an HTTPS endpoint, such as a Cloud Run service), or export (to BigQuery or Cloud Storage). Message retention, ordering, and dead-letter topics (where undeliverable messages go after repeated failures) are configured on the subscription. Pub/Sub is global: a topic is not tied to a region, though message storage policies can restrict where messages are stored for compliance.
Dataflow is the managed service for batch and streaming data processing, based on Apache Beam. A Dataflow job reads from a source (such as Pub/Sub, Cloud Storage, or BigQuery), transforms the data, and writes it to a destination. For the exam, the key points are that Dataflow is the answer for data-processing pipelines (ETL), that it autoscale worker VMs based on the workload, and that its jobs are monitored in the console where you can see throughput, watermarks, and worker status. Dataflow is a processing engine, not a storage service — it moves and transforms data between the storage services.
Loading data
Getting data into Google Cloud is a deployment task in its own right. The methods the exam guide lists are: command-line upload (for example, gcloud storage cp or gsutil cp to copy files into a Cloud Storage bucket), loading from Cloud Storage into a database or warehouse (for example, a BigQuery load job reading CSV or JSON files from a bucket, or a Cloud SQL import from a bucket), and the Storage Transfer Service for large or recurring transfers — from another cloud provider (such as Amazon S3), from an on-premises system, or between buckets. Storage Transfer Service is the answer when the transfer is large, recurring, or from another cloud; a simple gcloud storage cp is the answer for a one-off upload of a few files. For databases, BigQuery load jobs and Cloud SQL imports both read from Cloud Storage, so the pattern is usually "upload to a bucket first, then load from the bucket".
High availability and backup defaults
Availability and backup settings are chosen at deployment time, and the exam tests whether you know the defaults — and when to change them. Cloud SQL instances are zonal by default; enabling high availability (regional) adds a synchronous standby in a second zone with automatic failover, roughly doubling the instance cost. Automated backups are configurable at creation and should be enabled for any instance holding real data, with point-in-time recovery for the ability to restore to a specific moment. Firestore databases are regional or multi-regional at creation; multi-regional gives higher availability for globally accessed data. Spanner instances are regional or multi-regional by configuration, and multi-regional Spanner is the highest-availability relational option Google Cloud offers. BigQuery is serverless and multi-zonal by design — there is no availability setting to choose, because Google manages replication internally.
The backup story varies by service, and the exam's trap is assuming backups are automatic everywhere. Cloud SQL automated backups must be enabled (they are on by default for new instances in the console, but the setting is worth verifying). Firestore backups are scheduled explicitly, and point-in-time recovery is a separate setting. BigQuery tables can be protected with snapshots and time travel (querying a table as it was up to seven days ago), but there is no traditional backup schedule — the protection model is different. The principle for the exam: when a question mentions recovery, check which service it is about, and remember that the backup mechanism is per-service, not universal.
Choosing regions for data
Data services are region-scoped, and the region choice has three consequences the exam tests. The first is latency: the database should be in the same region as the application that uses it, because cross-region queries add latency to every request. The second is cost: data moving between regions is charged as egress, so a database in one region serving an application in another costs more than a co-located pair. The third is compliance: some data must stay within a particular geography, and the region (or multi-region) choice is how you enforce that. A common exam scenario is an application in europe-west2 with a database in us-central1: the fix is to move the database (or create a replica) in the application's region. For globally distributed applications, the answer is usually a multi-region or globally distributed service — Spanner, multi-region Firestore, or Cloud SQL with cross-region read replicas — rather than a single-region instance.
Exam traps to watch for
- Choosing a zonal Cloud SQL instance for a workload that must survive a zone failure: regional (HA) is the answer.
- Forgetting that a BigQuery dataset's location cannot be changed after creation.
- Using a one-off copy command for a large recurring transfer from another cloud: Storage Transfer Service is built for that.
- Confusing Pub/Sub push and pull: push delivers to an HTTPS endpoint; pull requires the subscriber to request messages.
Practice questions
Test yourself on this guide. Answers are checked instantly, with a full explanation for each question.
Q1.A company needs a Cloud SQL for PostgreSQL instance that stays available if a zone fails. Which configuration should they choose?
Explanation: A regional Cloud SQL instance keeps a synchronous standby in a second zone and fails over automatically if the primary zone fails. A zonal instance — even with backups or point-in-time recovery — goes down with its zone; backups help you recover data, they do not keep the service running. A read replica serves read traffic and does not automatically take over writes.
Exam objective: ACE exam guide 2.2: Planning and implementing storage and data solutions — deploying data products (Cloud SQL)
Q2.A team needs to transfer 50 TB of data from an Amazon S3 bucket into Cloud Storage, and repeat the transfer weekly to keep the data in sync. What should they use?
Explanation: Storage Transfer Service is designed for large transfers from other clouds and on-premises sources, with scheduling for recurring syncs and built-in verification. A manual gcloud storage cp is impractical for 50 TB and does not schedule itself. A Cloud Function copying objects one by one is slow, unreliable at this scale, and expensive. BigQuery load jobs read from Cloud Storage, not directly from S3.
Exam objective: ACE exam guide 2.2: Planning and implementing storage and data solutions — loading data (Storage Transfer Service)
Q3.A developer creates a BigQuery dataset in the US multi-region, then realizes company policy requires the data to stay in the EU. What should they do?
Explanation: A BigQuery dataset's location is fixed at creation and cannot be changed — the only remedy is a new dataset in the right location and a data copy. There is no location setting to edit. Projects are not located in regions. BigQuery Data Transfer Service moves data between sources on a schedule; it does not relocate a dataset.
Exam objective: ACE exam guide 2.2: Planning and implementing storage and data solutions — deploying data products (BigQuery)
Q4.A streaming application publishes sensor data to Pub/Sub. A Cloud Run service should receive each message automatically over HTTPS as soon as it arrives. Which subscription type should you configure?
Explanation: A push subscription delivers messages to an HTTPS endpoint automatically — exactly what a Cloud Run service needs to receive messages as they arrive. A pull subscription requires the subscriber to request messages, which does not fit an automatic push model. Export and BigQuery subscriptions write messages to BigQuery, not to a service.
Exam objective: ACE exam guide 2.2: Planning and implementing storage and data solutions — deploying data products (Pub/Sub)
Q5.A company wants to run an ETL pipeline that reads streaming data from Pub/Sub, transforms it, and writes the results to BigQuery. Which service should run the pipeline?
Explanation: Dataflow is built for this: managed Apache Beam pipelines that read from sources such as Pub/Sub, transform the data, and write to destinations such as BigQuery, for both batch and streaming. Cloud Functions could process individual messages but is not a pipeline engine for streaming ETL. Cloud SQL is a database, not a processing service. Cloud Scheduler only triggers work on a timetable.
Exam objective: ACE exam guide 2.2: Planning and implementing storage and data solutions — deploying data products (Dataflow)
Q6.A team needs to load a 10 GB CSV file from a local machine into BigQuery for analysis. What is the simplest correct sequence?
Explanation: The standard pattern is to upload the file to Cloud Storage (with gcloud storage cp) and then load it into BigQuery with a load job — BigQuery reads directly from buckets. Cloud SQL is the wrong destination for an analytics file. Publishing rows as Pub/Sub messages is absurd for a 10 GB file. Running the analysis on a VM ignores BigQuery entirely.
Exam objective: ACE exam guide 2.2: Planning and implementing storage and data solutions — loading data (command line upload, load data from Cloud Storage)
Q7.A developer is creating a Firestore database for a mobile app. Which statement about Firestore modes is correct?
Explanation: Firestore is created in either Native mode (the document database with real-time sync) or Datastore mode (for compatibility with the older Datastore), and the choice is permanent. Native mode is the right choice for a new mobile app. The mode cannot be switched after creation.
Exam objective: ACE exam guide 2.2: Planning and implementing storage and data solutions — deploying data products (Firestore)
Q8.A PostgreSQL workload on Cloud SQL needs significantly higher throughput and lower latency than Cloud SQL provides, with PostgreSQL compatibility preserved. Which service should the team evaluate?
Explanation: AlloyDB is Google's PostgreSQL-compatible database designed for demanding workloads that outgrow Cloud SQL, with higher performance and availability. Spanner has a PostgreSQL interface option but is a different, globally distributed database — not the direct Cloud SQL upgrade path. Bigtable is NoSQL and not PostgreSQL-compatible. Firestore is a document database, not a relational PostgreSQL replacement.
Exam objective: ACE exam guide 2.2: Planning and implementing storage and data solutions — deploying data products (AlloyDB)