GCP ACE Practice / Study guide / Topics / Choosing the Right Compute Option
Choosing the Right Compute Option
What actually gets tested on the Google Cloud Associate Cloud Engineer exam,
the traps to know, and 6 free sample questions with full explanations.
This is a pure decision-framework topic, and the exam rewards recognizing the scenario's keywords over knowing every product's feature list. Compute Engine is the answer whenever you see full OS control, custom agents, specialized hardware, or a lift-and-shift migration — you own the operating system, with everything that implies. Serverless platforms (Cloud Run, Cloud Functions, App Engine standard) abstract the OS away entirely; whenever a scenario says "minimize operational overhead" or "no infrastructure to manage," the answer is one of these three, not Compute Engine or GKE.
Within serverless, the differentiator is trigger shape: Cloud Run runs containers that scale from zero based on request load with no infrastructure to manage — the general-purpose serverless answer for anything that fits in a container. Cloud Functions is the textbook choice for event-driven, short-lived, per-object work (a Cloud Storage upload invoking a function directly via Eventarc is the classic exam scenario); polling for events instead of reacting to them directly is a frequent wrong-answer setup.
Between the two container-orchestration options, GKE Autopilot is full Kubernetes with Google operating the nodes (you pay per pod, keep the K8s API surface), GKE Standard leaves node management to you for more control, and Cloud Run hides Kubernetes entirely for teams that just want a container to run — reach for GKE specifically when a scenario needs Kubernetes-native features (custom schedulers, specific networking/CRDs), and Cloud Run when it just needs "run this container."
Exam tips for this topic
- "Minimize ops overhead / no infrastructure to manage" points to Cloud Run, Cloud Functions, or App Engine — never Compute Engine or GKE Standard.
- Event-driven, short-lived, per-object triggers (e.g. a Cloud Storage upload) point to Cloud Functions via Eventarc, not a polling loop on a VM.
- Reach for GKE specifically when a scenario needs Kubernetes-native features; reach for Cloud Run when it just needs to run a container without touching K8s at all.
Sample questions — Choosing the Right Compute Option
These 6 questions are drawn from our full bank of 658 to show the style and
depth you'll get in the app. Each comes with the full explanation you'd see after answering
in a real practice session.
Compute choice
You need to run a stateless containerized HTTP API that scales to zero when idle and requires no cluster management. Which service fits best?
- Cloud RunCorrect
- Compute Engine managed instance group
- GKE Standard
- Bare metal solution
Why: Cloud Run runs containers serverlessly, scales from zero to many instances per request load, and needs no infrastructure management. MIGs and GKE both leave you managing VMs/clusters and don't scale to zero (GKE Autopilot pods still need the cluster); bare metal is the opposite of the requirement.
Compute choice
A team must lift-and-shift a legacy Windows application that needs full OS control and a specific licensed agent installed. Which compute option?
- Compute Engine VMCorrect
- Cloud Run
- Cloud Run functions
- App Engine standard
Why: Full OS control, custom agents, and lift-and-shift map to Compute Engine — you own the OS. Serverless platforms (Cloud Run, functions, App Engine standard) abstract the OS away and can't host arbitrary Windows services.
Compute choice
Your microservices need Kubernetes features (custom operators, DaemonSets, service mesh) but the team wants Google to manage nodes, scaling, and security patching. What do you choose?
- GKE AutopilotCorrect
- GKE Standard
- Cloud Run
- Unmanaged instance group
Why: Autopilot is full Kubernetes with Google operating the nodes — you pay per pod and keep K8s APIs. Standard leaves node management to you; Cloud Run hides Kubernetes entirely, losing the K8s-specific features required here.
Compute choice
You need code to run for ~2 seconds in response to each new file uploaded to a Cloud Storage bucket. The simplest fit is:
- A Cloud Run function triggered by the bucket's finalize eventCorrect
- A small VM that polls the bucket for new objects in a loop
- A GKE cluster running a custom controller that watches the bucket
- A Cloud SQL trigger fired when the object metadata row is inserted
Why: Event-driven, short-lived, per-object work is the textbook serverless function use case — the storage event invokes the function directly (via Eventarc). Polling wastes money, GKE is heavy machinery for 2 seconds of work, and Cloud SQL triggers don't respond to bucket events.
Compute choice
An App Engine standard app is a good fit when: (Choose two.)
- Traffic is spiky and scaling to zero between requests saves moneyCorrect
- The app is a supported-runtime web app needing zero server managementCorrect
- The app needs to write files to its local disk and keep them permanently
- The app requires a custom kernel module loaded at boot time
Why: App Engine standard runs supported language runtimes in a sandbox with request-based autoscaling to zero. The sandbox has no durable local disk and certainly no kernel access — those needs point to Compute Engine (or flexible/Cloud Run for containers).
Compute choice
A monthly batch job needs 96 vCPUs for 3 hours; the rest of the month, nothing. The most cost-effective design is:
- Create instances (ideally Spot) on demand for the run, then delete themCorrect
- Keep a 96-vCPU VM running all month with committed use discounts
- Buy a sole-tenant node so the job always has dedicated capacity
- Run the job in Cloud Shell, which is free for up to 50 hours a week
Why: Cloud economics reward ephemeral usage: create → run 3h → delete costs ~1/240th of an always-on VM, and Spot cuts that further for interruptible work. CUDs and sole tenancy commit you to constant capacity; Cloud Shell isn't for heavy compute.
Practice this for real
The app has dozens more questions on this and every other topic, with instant scoring,
spaced repetition of what you get wrong, and progress tracking across sessions.
Start practicing →
This topic shows up in
Other topics