| name | gcp-patterns |
| description | GCP Patterns: Cloud Functions, Cloud Run, BigQuery, Pub/Sub, GKE, serverless data. |
| triggers | {"extensions":[".py",".ts",".yaml"],"directories":["gcp/","infrastructure/"],"keywords":["gcp","google cloud","cloud functions","cloud run","bigquery","pubsub","gke","dataflow","cloud build"]} |
| auto_load_when | Building on Google Cloud or designing serverless architectures |
| agent | cloud-architect |
| tools | ["Read","Write","Bash"] |
GCP Architecture Patterns
Focus: Serverless, data analytics, managed services
1. Compute Selection
When to use what:
├── Cloud Functions (2nd gen)
│ ├── Event-driven, HTTP, scheduled
│ ├── V2: longer timeout (60→9h), more memory (4GB→32GB)
│ └── Pay per invocation + compute time
│
├── Cloud Run
│ ├── Container-based, HTTP services
│ ├── Any language, any framework
│ ├── Auto-scales to 0, scales to 1000+ replicas
│ └── Pay per request (100ms minimum)
│
├── GKE (Kubernetes)
│ ├── Complex orchestration
│ ├── Multi-service architecture
│ └── Full Kubernetes ecosystem
│
└── Compute Engine
│ ├── Legacy, specific hardware
│ └── VMs with full control
2. Serverless Data Patterns
GCP Data Services:
├── BigQuery
│ ├── Serverless data warehouse
│ ├── Petabyte scale, pay per query
│ └── ML with BigQuery ML
│
├── Dataflow
│ ├── Unified batch/streaming
│ ├── Apache Beam-based
│ └── Auto-scaling, no cluster management
│
├── Pub/Sub
│ ├── Real-time messaging
│ ├── Global (publish once, subscribe regionally)
│ └── At-least-once delivery
│
├── Dataproc
│ ├── Spark/Hadoop clusters
│ ├── On-demand, auto-scales
│ └── Batch processing
│
└── Cloud Storage
├── Object storage, multiple classes
└── Lifecycle policies, uniform buckets
3. Serverless ML
GCP AI/ML:
├── Vertex AI
│ ├── End-to-end ML platform
│ ├── AutoML for quick start
│ └── Custom training with distributed
│
├── Prediction endpoints
│ ├── Real-time prediction (online)
│ ├── Batch prediction (offline)
│ └── Vertex AI endpoints (managed)
│
├── Feature Store
│ ├── Centralized feature registry
│ ├── Online/offline serving
│ └── Feature sharing across models
│
└── Model Registry
├── Version control for models
├── A/B testing, canary deployment
└── Audit trail
4. Networking & API
GCP Networking:
├── API Gateway
│ ├── Serverless API management
│ ├── Rate limiting, authentication
│ └── Integrates with Cloud Functions/Run
│
├── Cloud Load Balancing
│ ├── Global HTTP(S) load balancer
│ ├── SSL termination, CDN
│ └── Traffic management (routing rules)
│
├── VPC
│ ├── Shared VPC for organization
│ ├── Private Google Access
│ └── VPC flow logs
│
└── Cloud CDN
├── Global CDN
└── Signed URLs/cookies for private content
5. Infrastructure as Code
IaC in GCP:
├── Deployment Manager
│ ├── GCP native, YAML-based
│ └── Good for simple stacks
│
├── Terraform (recommended)
│ ├── Multi-cloud, large ecosystem
│ └── Official Google provider
│
└── Config Connector
├── Kubernetes-style for GKE
└── Manage GCP resources via K8s API
Key Patterns
- Serverless by default - Cloud Functions → Cloud Run → GKE
- Data-first - BigQuery as data platform
- Managed services - Prefer managed over self-hosted
- Multi-region - Global services, regional deployment
- Cloud Armor - Security at edge
Anti-Patterns
❌ Using Compute Engine for everything
✅ Serverless (Functions/Run) for most workloads
❌ No VPC for serverless services
✅ Serverless VPC Access Connector for private networking
❌ Not using Cloud CDN
✅ Enable for all static assets
❌ No API management
✅ API Gateway or Apigee for production APIs
❌ Ignoring BigQuery
✅ It's serverless, cheap, use for analytics unless specific reason not to
Quick Reference
| Service | Use Case | Key Feature |
|---|
| Cloud Functions | Event-driven | Pay per invoke |
| Cloud Run | Container services | Any language |
| BigQuery | Data warehouse | Serverless SQL |
| Pub/Sub | Messaging | Global |
| GKE | Kubernetes | Managed K8s |
| Cloud Storage | Object storage | Classes |
| Vertex AI | ML platform | AutoML |
| Cloud CDN | CDN | Global |
🌍 Universal Language Support
- Turkish Native: This skill natively supports Turkish. If the user prompt is in Turkish, all analysis, formatting, and output MUST be entirely in Turkish. You do not need explicit "write in Turkish" instructions.
