Google Cloud ACE Cheat Sheet: Core Services, IAM, and Deployments

Google Cloud ACE cheat sheet covering compute, storage, IAM, networking, deployment, and operations.

Use this as a last-mile review page. ACE gets much easier when you stop treating every question as a compute question and classify it in this order:

  1. Scope: organization, folder, project, region, zone, resource, or principal?
  2. Service fit: runtime, data product, network entry point, or storage/database lane?
  3. Control boundary: IAM, service account, subnet, route, firewall, or peering?
  4. Operations signal: deployment, logs, metrics, audit, backup, snapshot, or rollback?

ACE answer sequence

Use this when the stem mixes scope, service fit, control boundary, or operations evidence.

    flowchart TD
	  S["Scenario"] --> C["Classify scope first"]
	  C --> F["Pick the service fit"]
	  F --> B["Check the control boundary"]
	  B --> O["Check the operations signal"]

Fastest 10-minute review

If the question says… Strongest first lane
business unit, billing separation, or API enablement project or folder boundary
“simple containerized app” with minimal ops Cloud Run before heavier runtimes
cluster primitives, node pools, or Kubernetes control GKE
outbound internet from private VMs Cloud NAT, not public IPs on every VM
“who changed this” or “who accessed this” audit logs
threshold alert, SLO drift, or CPU saturation Cloud Monitoring
app event detail or request failure trace Cloud Logging
team or workload needs access IAM role, binding, scope, or service account before network guesses

ACE review map

Official section Weight Use this cheat-sheet section first
1. Setup ~20% scope ladder, project and billing rules, IAM basics
2. Planning ~17.5% runtime picker, storage/database picker, network chooser
3. Deployment ~25% deployment lanes, VPC rules, service composition
4. Operations ~20% logging vs monitoring, recovery and rollback cues
5. Access & Security ~17.5% IAM scope, service accounts, least privilege, impersonation

Scope and control-first decision order

    flowchart TD
	  Org["Organization"] --> Folder["Folder"] --> Project["Project"] --> Place["Region / Zone"] --> Resource["Resource"]
	  Resource --> Control["Control Boundary"]
	  Control --> Service["Service Fit"]
	  Service --> Net["Network Path"]
	  Net --> Ops["Operations Signal"]

Scope ladder

Layer What it really answers Strong exam reading
organization top governance boundary org-wide guardrail or admin pattern
folder grouped projects under one branch business-unit or environment grouping
project billing, IAM, API enablement, and quota boundary the main ACE operating lane
region multi-zone placement area residency, latency, and resilience choice
zone single failure-domain placement VM or zonal resource location
resource one object or service instance the exact thing being changed

Scope traps

If the stem says… Do not jump to… Stronger first reading
“team needs access” bigger VM or another subnet IAM member, role, and scope
“charge this to a different unit” resource labels only project or billing-account structure
“users in multiple regions need the same entry path” regional-only thinking global service or multi-region design
“private VM cannot reach a service” immediate restart or redeploy route, firewall, NAT, DNS, or peering boundary

Runtime and service picker

Requirement Strongest first lane Common trap
full OS control, legacy software, or custom agents Compute Engine choosing VMs when a managed runtime is enough
containerized app with the least platform work Cloud Run expecting it to replace every cluster need
Kubernetes APIs, node pools, or cluster policy GKE choosing it without a real Kubernetes requirement
event-triggered short logic Cloud Functions stretching it into full app hosting
large-scale analytical SQL BigQuery treating Cloud SQL like a warehouse
object durability, exports, archives, and media Cloud Storage confusing it with attached disks
transactional relational application store Cloud SQL using it for warehouse-style analytics
globally distributed document pattern Firestore forcing relational assumptions onto it
event fan-out and async messaging Pub/Sub using it as a database

Deployment and change-control cues

If the question is really about… Strongest first lane
repeatable infrastructure review and rollout Terraform or IaC workflow
container revision traffic shaping Cloud Run rollout or traffic split
cluster deployment primitives GKE + kubectl, manifests, or Helm
one-off admin shell work Cloud Shell or gcloud
image-based VM fleet standardization instance templates and managed instance groups

Storage and database fit

Requirement Strongest first lane Why
immutable files, backups, exports, and object lifecycle Cloud Storage storage classes and object semantics
petabyte analytics over large datasets BigQuery warehouse and analysis path
managed MySQL, PostgreSQL, or SQL Server Cloud SQL operational relational workload
globally distributed JSON-like document access Firestore document database path
attached block storage for one VM or instance template persistent disk block semantics
managed shared file path Filestore file protocol and shared-mount path

High-confusion storage pairs

Pair Keep this distinction clear
Cloud Storage vs persistent disk object storage versus attached block storage
BigQuery vs Cloud SQL analytics warehouse versus transactional relational engine
Cloud SQL vs Firestore relational transactions versus document model
storage class choice vs engine choice cost-retention lever versus application data-model decision

Networking and traffic path rules

Requirement Strongest first lane
explicit subnet planning custom mode VPC
default starter subnet layout auto mode VPC
permit or block traffic firewall rules or higher policy controls
private traffic between cloud networks peering or the right private connectivity model
private outbound internet from internal VMs Cloud NAT
internal or external name resolution Cloud DNS
global entry path to distributed backends global load balancing
regional entry path regional load balancing

Network troubleshooting order

Check Why it comes early
DNS bad names look like bad routes
route no path means no connectivity
firewall reachable networks can still block traffic
NAT or egress design private VMs often need outbound internet without public IPs
peering or private connectivity boundary the path itself may be missing

Networking traps

Trap Better reading
“peering fixes everything” peering gives path only; routes and firewalls still matter
“VM needs internet” first ask whether it needs public inbound reachability or only outbound egress
“load balancer” first classify global versus regional
“DNS issue” separate naming from routing and filtering

IAM and security quick rules

If the question is really about… Strongest first lane
who has access right now inspect IAM policy at the relevant scope
common job-function permissions predefined role
narrower least-privilege gap custom role
workload identity without static user credentials service account
temporary delegated use of another principal service account impersonation
broad governance guardrail higher-scope organization or folder policy

IAM boundary table

Item What it is Do not confuse it with
role permission bundle a principal
binding assignment of role to member the role definition itself
service account workload principal a human administrator
impersonation short-lived delegated use distributing long-lived keys
audit log record of who did what performance metrics

Operations and recovery cues

Symptom or need Inspect first
thresholds, CPU, latency, saturation, uptime trend Cloud Monitoring
request failures, app events, stack traces, detailed system messages Cloud Logging
admin activity or access history audit logs
bad deployment version revision history, rollout controls, or traffic split
VM recovery path snapshots, images, backups, or instance history
private service reachability failure route, firewall, NAT, subnet, and target location

Logging versus monitoring

Question Cloud Logging Cloud Monitoring
Main job event records and detailed log messages metrics, dashboards, thresholds, and alerts
Strongest first when you need what happened you need trend or health signal
Common trap expecting logs alone to act like alerts expecting metrics to explain the full event record

High-confusion pairs

Pair Keep this distinction clear
project vs folder workload boundary versus higher admin grouping
region vs zone broader placement area versus one failure-domain location
role vs binding permission set versus assignment
service account vs user workload identity versus human identity
Cloud Run vs GKE managed container runtime versus Kubernetes control need
logging vs audit general event record versus governance-focused who-did-what record
peering vs Cloud NAT private network reachability versus outbound internet egress
global vs regional load balancing cross-region entry versus region-scoped entry

Operator tools to recognize quickly

Tool or surface Why ACE mentions it
gcloud primary CLI for create, inspect, and change operations
Cloud Shell fast browser-based admin shell with Google Cloud tooling
kubectl cluster and workload operations on GKE
Terraform repeatable reviewed infrastructure changes
Cloud Console visual admin and inspection path

Last 15-minute review

Recheck this Because the miss often hides here
scope words such as project, folder, region, zone, and principal wrong answers often solve the wrong boundary
managed versus self-managed runtime fit the simpler managed service usually wins
logging versus monitoring versus audit these are different operator lanes
route, firewall, NAT, and peering networking questions punish blurred boundaries
role, binding, service account, and impersonation IAM questions often fail on one boundary mismatch

What strong ACE answers usually do

  • fix the scope or access boundary before tuning the workload
  • choose the most operationally appropriate managed service
  • separate private connectivity, traffic filtering, and name resolution
  • inspect the right signal before making a broad infrastructure change
  • prefer the answer that is easier to operate, audit, and explain
Revised on Sunday, May 10, 2026