Managing Storage, Databases & Networking

Google ACE ACE · Ensuring the successful operation of a cloud solution

By Abu Anns · Last updated 20 November 2026Exam version: ACE (exam guide live 30 June 2026)

Operating storage and databases

Once storage and databases are deployed, the operations work is keeping them secure, cost-effective, and recoverable. For Cloud Storage that means managing who can access objects, automating lifecycle transitions, and controlling costs. For databases it means running queries, estimating storage costs, and backing up and restoring instances. For networking it means growing subnets, reserving IP addresses, and managing DNS and NAT. These tasks are the everyday work of the operations section, and the exam tests them as concrete procedures.

Managing Cloud Storage objects

Access to objects in a bucket is controlled in two ways. Uniform bucket-level access (the recommended setting) means IAM roles on the bucket decide who can read or write all objects in it — one policy, consistently applied. Fine-grained access (the legacy mode) allows per-object ACLs, where each object can have its own access list. Uniform access is simpler, auditable, and the exam's preferred answer whenever consistent access control is mentioned; fine-grained access is only needed when individual objects genuinely need different permissions. Buckets can also be made public, but that is a deliberate, visible choice — the console warns about it — and it is almost never the right answer in an exam scenario.

Object Lifecycle Management automates what happens to objects as they age. Lifecycle rules have a condition (for example, "objects older than 30 days", or "objects with a particular prefix") and an action: transition the object to a colder storage class (Nearline, Coldline, Archive), or delete it. The classic pattern is a bucket of logs or uploads that transitions to Nearline after 30 days, Coldline after 90, and is deleted after a year. Lifecycle rules save money automatically and enforce retention policy without manual work. Note that lifecycle transitions respect minimum storage durations: moving an object to Archive and deleting it before 365 days incurs an early-deletion charge.

Managing databases: queries, costs, backups

The exam guide lists executing queries to retrieve data from Cloud SQL, BigQuery, Spanner, Firestore, and AlloyDB. In practice this means using the right tool for each service: the BigQuery console or bq query for BigQuery, the Cloud SQL console or a database client for Cloud SQL, and the Firestore console or client libraries for Firestore. The exam does not test SQL syntax in depth; it tests knowing which tool queries which service, and that BigQuery queries are charged by the data processed.

Estimating costs of data storage resources is a planning task that recurs in operations: the Google Cloud Pricing Calculator gives estimates for storage, database instances, and query costs, and the console's billing reports show actuals. For BigQuery, the cost lever is the amount of data scanned — partitioning and clustering tables reduces the data each query reads. For Cloud SQL, the levers are instance size, storage size, and backup retention.

Backing up and restoring differs by service. Cloud SQL supports automated daily backups with point-in-time recovery, and on-demand backups taken at any time; a restore creates a new instance from the backup (or restores the same instance to an earlier point). Firestore supports scheduled backups and point-in-time recovery, with exports to Cloud Storage for longer-term retention. The exam point is that backups must be configured — they are not automatic in every service — and that restore is a tested procedure, not a theoretical one. Reviewing job status is the last item: Dataflow jobs and BigQuery jobs are monitored in their consoles, where you can see progress, throughput, and errors, and cancel a runaway job.

Managing networking: subnets, IPs, DNS, NAT

Network operations tasks include adding a subnet to an existing VPC (choose a region and a non-overlapping IP range), and expanding a subnet to have more IP addresses (widening its CIDR range — possible because subnet ranges can grow, but never shrink, and the expansion must not overlap another subnet's range). Both are routine as an environment grows.

Reserving static IP addresses gives a VM or load balancer an address that does not change. External static IPs are reserved for resources that users or partners whitelist; internal static IPs are reserved for VMs that other resources address by IP. An ephemeral IP, by contrast, is assigned at creation and can change if the resource is stopped and restarted — the exam's trap is assuming an ephemeral address is stable. Unused static IPs incur a small charge, so reserving addresses you do not use wastes money.

Cloud DNS is Google's managed DNS service: it hosts DNS zones (public zones for internet-facing names, private zones visible only inside a VPC) and answers queries with low latency worldwide. Private DNS zones are how internal services get stable names inside a VPC without exposing them publicly. Cloud NAT lets VMs without external IP addresses reach the internet for outbound traffic (for example, to download OS updates) while remaining unreachable from the internet — the NAT gateway translates the VMs' private addresses to a shared public address for outgoing connections only. The exam pairing is common: private VMs plus Cloud NAT is the secure pattern for VMs that need outbound internet access but must not be internet-reachable.

Reviewing job status: Dataflow and BigQuery

Data processing jobs are long-running operations, and checking on them is part of the daily routine. Dataflow jobs are monitored in the Dataflow console, where each job shows its status (running, succeeded, failed, cancelled), its throughput (elements processed per second), its watermark (how far behind real time a streaming job is), and the number of workers it is using. A streaming job whose watermark is falling behind is not keeping up with its input — the fix is usually more workers or a less expensive transform. A failed job shows the error and the step that failed, and the job can be drained (for streaming jobs, letting in-flight data finish) or cancelled outright. BigQuery jobs — queries and load jobs — are listed in the BigQuery console's job history, where you can see how long each took, how much data it scanned, and whether it succeeded. A query that scanned far more data than expected is usually missing a partition filter, and the job history is where you spot it. Both consoles let you cancel a runaway job, which is the right response to a query that is scanning petabytes by mistake.

The exam tests job monitoring at a practical level: where to look, what the status means, and what to do about a problem. The key associations are simple — Dataflow jobs in the Dataflow console, BigQuery jobs in the BigQuery console — and the common thread is that both services expose enough detail to diagnose a slow or failing job without guessing. When a question describes a data pipeline that is behind schedule or a query that costs too much, the first step is always to check the job status and its metrics before changing anything.

Advertisement

Exam traps to watch for

Practice questions

Test yourself on this guide. Answers are checked instantly, with a full explanation for each question.

Q1.A company stores application logs in a Cloud Storage bucket. Logs older than 30 days are rarely read, and logs older than one year must be deleted. What should they configure?

Report an error

Q2.A bucket contains sensitive documents, and the security team wants a single, auditable access policy for all objects in it — no per-object exceptions. Which setting should they use?

Report an error

Q3.A Cloud SQL database was accidentally corrupted by a bad migration at 2:00 PM. The team needs the data as it was at 1:00 PM. Automated backups and point-in-time recovery are enabled. What should they do?

Report an error

Q4.A subnet in us-central1 is running out of IP addresses as more VMs are added. The subnet is 10.0.0.0/24. What is the correct fix?

Report an error

Q5.A partner company whitelists your application's IP address in their firewall. Your VM currently uses an ephemeral external IP. What should you do to make sure the address does not change?

Report an error

Q6.A set of VMs has no external IP addresses, but they need to download OS security updates from the internet. They must not be reachable from the internet. What should you configure?

Report an error

Q7.A team wants internal services to be reachable by stable DNS names (such as db.internal.example.com) inside their VPC, without exposing those names on the public internet. What should they use?

Report an error

Q8.A BigQuery analyst runs the same daily report query over a 5 TB table, and the query costs are high. What is the best way to reduce the cost of the query?

Report an error