GCP Professional Cloud Architect (PCA) - Exam Strategy¶
Format reminder¶
- 50-60 questions, 120 minutes
- Pass mark: not published; widely reported as ~70-75% based on community feedback
- Multiple choice + multiple response
- Includes case study questions referencing fictional companies (Mountkirk Games, EHR Healthcare, Helicopter Racing League, TerramEarth)
Top traps¶
- Spanner vs Cloud SQL vs Firestore vs BigQuery:
- Spanner: globally consistent SQL, horizontal scale, ACID across regions
- Cloud SQL: regional, managed PostgreSQL/MySQL/SQL Server
- Firestore: document store, regional or multi-region
-
BigQuery: serverless analytics warehouse Each has a specific use case; PCA tests the boundary cases.
-
GCE machine families: E2 (cost-optimized), N2 (balanced), C3 (latest compute-optimized), M3 (memory-optimized for in-memory DBs and SAP HANA). Pick by workload shape.
-
CUD vs SUD vs Spot:
- SUD: automatic, ~30% discount for sustained running
- CUD: 1-3 year commitment, up to 70% (resource-based or spend-based)
-
Spot (formerly preemptible): 60-91% discount, can be reclaimed Newer CUDs (flexible, spend-based) blur the line with Savings Plans on AWS.
-
Interconnect tiers:
- Dedicated Interconnect: 10 Gbps or 100 Gbps direct
- Partner Interconnect: 50 Mbps - 50 Gbps via SP
- HA VPN: over public internet, 99.99% with two tunnels
-
SLAs: 99.9% (single Interconnect) β 99.99% (two metros, single region) β 99.99% (dual region)
-
GKE Standard vs Autopilot: Standard for full control + node management; Autopilot for least-ops, pod-based pricing. PCA increasingly favors Autopilot for "minimal operational overhead."
-
VPC Service Controls vs Private Google Access vs Private Service Connect:
- Private Google Access: VMs without public IPs reach Google APIs
- VPC Service Controls: data exfiltration prevention via service perimeters
-
Private Service Connect: consume Google services or third-party services via private IPs
-
IAM hierarchy: Organization β Folders β Projects β Resources. Roles flow down. Org policies are constraints (different from IAM) - they restrict configurations.
-
Resource hierarchy and billing: each Project has its own quotas, billing, IAM. Folders for grouping. Org policy constraints at any level.
-
Cloud Storage classes: Standard, Nearline (>30 days), Coldline (>90 days), Archive (>365 days). Lifecycle policies for transitions. Multi-region vs Dual-region vs Region location selection.
-
Network tiers: Premium (default, Google's global network) vs Standard (cheaper, regional egress through public internet). Premium is the default modern answer.
Case study tactics¶
The PCA case studies are real - questions for each case study reference specific constraints in the company's Existing Environment and Business Requirements sections. Read the case study once carefully; for each subsequent question, jump back to the relevant constraint.
Common case study traps: - "Existing Java app" - rules out Cloud Functions (Python/Node/Go runtime, though Java exists in Functions Gen2) - "Cannot move data outside the EU" - rules out multi-region us / global Spanner - "Must minimize cost" - prefer Cloud Run over GKE, Cloud Functions over Cloud Run for spiky - "Must minimize operational overhead" - prefer fully-managed (Cloud Run, BigQuery, Cloud SQL) over self-managed
Time management¶
120 / ~55 = ~2.2 min/question. Pace: half done by minute 60. Leave 15 min for flagged review.
When stuck¶
- Identify case study constraints that filter options.
- Default to managed > self-managed when both work.
- Default to Google's recommended pattern - Cloud Service Mesh, GKE Autopilot, Spanner for global, Cloud Run for serverless containers.
- Eliminate "build it yourself" if a Google-managed service does it.
Day-of logistics¶
120 min, ~55 questions: bring two IDs. Online proctored available via Kryterion / Pearson VUE.
After¶
Pass: Cert valid 2 years. Renew by passing again or another GCP Professional cert.
Fail: Most failures are on case study questions where missed constraints cost points. Re-read the case studies in the official guide and practice mapping requirements to GCP services.
PCA patterns¶
- "Globally consistent SQL with sub-second writes" = Cloud Spanner
- "Multi-region serverless containers" = Cloud Run + Global LB with serverless NEGs
- "Hybrid 99.99%" = two Dedicated Interconnects in two metros
- "GKE with mTLS + observability + minimal ops" = Autopilot + Cloud Service Mesh
- "Steady 24/7 cost optimization" = 3-year CUDs
- "Interruptible cheap compute" = Spot
- "Data exfil prevention" = VPC Service Controls
- "Multi-tenant SaaS data isolation in BigQuery" = dataset IAM + row-level + column policy tags
- "Cross-region DR for stateful app" = Cloud SQL cross-region read replica + MIG scale-to-zero in DR
- "Per-call zero trust between services" = Cloud Service Mesh + AuthorizationPolicy