Study Google Cloud ACE Compute Choices: key concepts, common traps, and exam decision cues.
This lesson is about matching the workload to the right compute model. ACE often asks whether the best answer is a VM, a managed container service, a serverless runtime, or a lower-cost VM option.
Spot VM: Deeply discounted VM option that can be interrupted, making it better for fault-tolerant or disposable workloads than critical always-on workloads.
Custom machine type: Compute Engine VM shape tailored to a workload’s CPU and memory ratio rather than using a fixed predefined type.
Cloud Run functions: Current Google Cloud functions path for lightweight event-driven code. Older material may still say Cloud Functions.
Google Cloud wants you to separate:
| Requirement | Strong lane |
|---|---|
| fine-grained OS control or legacy app behavior | Compute Engine |
| Kubernetes-managed container platform | GKE |
| fully managed container runtime with simpler operations | Cloud Run |
| event-driven lightweight function execution | Cloud Run functions |
| interrupt-tolerant cost-sensitive compute | Spot VMs |
| If the question says | Strongest first lane | Why |
|---|---|---|
| legacy software, startup scripts, custom OS packages, direct VM control | Compute Engine | You need VM-level control |
| Kubernetes objects, cluster operations, node pools, container orchestration | GKE | This is Kubernetes-managed infrastructure |
| deploy a container without managing a cluster | Cloud Run | Managed container runtime |
| single-purpose event handler or lightweight triggered function | Cloud Run functions | Function-shaped event response |
| cheap compute for fault-tolerant batch or disposable jobs | Spot VMs | Lower cost in exchange for interruption risk |
flowchart TD
A["Need to run application code"] --> B{"How much platform control?"}
B -->|VM or OS control| C["Compute Engine"]
B -->|Kubernetes control plane| D["GKE"]
B -->|Managed container without cluster ops| E["Cloud Run"]
B -->|Single-purpose event-driven function| F["Cloud Run functions"]
C --> G["Use Spot VMs only if interruption is acceptable"]
ACE is usually not testing deep machine-family memorization. It is testing whether you notice when the problem is really:
| Situation | Strongest first move |
|---|---|
| workload needs a nonstandard CPU-to-memory ratio | Consider a custom machine type |
| workload is stateful or critical and cannot disappear suddenly | Avoid Spot VMs as the primary answer |
| workload is batch, retryable, or disposable | Spot VMs become realistic |
| Trap | Better reading |
|---|---|
| “Containers always mean GKE.” | If the question wants containers without cluster administration, Cloud Run is often stronger. |
| “Serverless code means Cloud Run only.” | Event-driven function prompts often point to Cloud Run functions. |
| “Lowest cost means Spot VMs no matter what.” | Spot VMs are only strong when interruption is acceptable. |
| “Custom machine type means a different product family.” | It is still Compute Engine, just a tailored VM shape. |
A team needs to deploy a containerized internal API. They do not want to manage Kubernetes, and the service must stay available under normal production traffic. Another batch job in the same company is retryable and mainly cost-sensitive.
The strongest reading is:
Correct answer: 1. The service-shaped container fits Cloud Run, while the retryable batch job can accept Spot VM interruption.