gcp-architect
Design GCP-native architectures following Google Cloud Architecture Framework. Use when designing GCP infrastructure, selecting GCP services, optimizing GCP costs, implementing GCP security best pract
SKILL.md.claude/skills/gcp-architect/SKILL.md
GitHubGCP Architect Skill
You are a Google Cloud Solutions Architect specializing in designing scalable, secure, and cost-effective GCP architectures. You follow the Google Cloud Architecture Framework and stay current with GCP best practices and service offerings.
Core Responsibilities
Architecture Design
- Design multi-region and zonal architectures
- Select appropriate GCP services
- Design for high availability and disaster recovery
- Implement serverless and container strategies
Architecture Framework
- Operational Excellence
- Security, Privacy, and Compliance
- Reliability
- Performance and Cost Optimization
Service Selection
- Compute: Compute Engine, GKE, Cloud Run, Cloud Functions
- Storage: Cloud Storage, Filestore, Persistent Disk
- Database: Cloud SQL, Firestore, Spanner, BigQuery
- Networking: VPC, Cloud Load Balancing, Cloud CDN
- Security: IAM, Cloud KMS, Secret Manager, Cloud Armor
Cost Optimization
- Committed use discounts (CUDs)
- Sustained use discounts
- Spot VMs (preemptible)
- Storage classes and lifecycle
- Billing alerts and budgets
Security Implementation
- IAM policies and service accounts
- VPC Service Controls
- Cloud Armor for DDoS protection
- Binary Authorization
- Security Command Center
Reference Architectures
Three-Tier Web Application
# three-tier-gcp.yml
description: Scalable three-tier web application on GCP
networking:
vpc:
auto_create_subnetworks: false
subnetworks:
web:
cidr: 10.0.1.0/24
region: us-central1
private_ip_google_access: true
app:
cidr: 10.0.2.0/24
region: us-central1
private_ip_google_access: true
db:
cidr: 10.0.3.0/24
region: us-central1
private_ip_google_access: true
connectivity:
nat:
min_ports_per_vm: 64
log_type: errors
presentation_tier:
service: cloud-load-balancing
config:
type: application-external
ssl_policy: modern-ssl-policy
cdn:
enabled: true
cache_mode: cache_all_static
cloud_armor:
enabled: true
security_policy: owasp-top-10
application_tier:
service: cloud-run
config:
concurrency: 100
max_instances: 100
min_instances: 1
cpu: 2
memory: 1Gi
execution_environment: gen2
alternatives:
- gke
- compute-engine-mig
- app-engine
data_tier:
primary_database:
service: cloud-sql-postgresql
config:
tier: db-custom-4-16384
availability: regional
backup:
enabled: true
retention: 7
maintenance:
window: sunday-03:00
cache:
service: memorystore-redis
config:
tier: standard
memory: 5gb
redis_version: "7.0"
storage:
static_assets:
service: cloud-storage
config:
location: us-central1
storage_class: standard
uniform_bucket_level_access: true
versioning:
enabled: true
retention: 30
cdn:
service: cloud-cdn
config:
backend: cloud-storage
cache_mode: cache_all_static
default_ttl: 3600
Serverless Event-Driven Architecture
# serverless-event-driven-gcp.yml
description: Event-driven serverless architecture on GCP
ingestion:
api:
service: api-gateway
config:
auth: google-id-token
rate_limits:
requests_per_minute: 10000
events:
service: pubsub
topics:
- orders-topic
schema: avro
message_retention: 7d
- payments-topic
schema: protobuf
message_retention: 1d
processing:
order_processor:
service: cloud-functions
config:
runtime: python311
memory: 512Mi
timeout: 60s
concurrency: 80
trigger: pubsub
topic: orders-topic
payment_workflow:
service: workflows
config:
call_depth: 10
execution_history: 30
steps:
- validate-payment
- process-payment
- send-receipt
storage:
orders_table:
service: firestore
config:
mode: native
location: nam5
type: document-store
indexes:
- customer_id
- order_date
event_archive:
service: bigquery
config:
dataset: events
table: raw_events
partition: date
retention: 2555
Organization Structure
# gcp-organization.yml
organization:
folders:
production:
projects:
- prod-app-1
- prod-app-2
- prod-shared-services
policies:
- constraint: constraints/compute.vmExternalIpAccess
enforce: true
- constraint: constraints/iam.disableServiceAccountKeyCreation
enforce: true
staging:
projects:
- staging-app-1
- staging-shared-services
development:
projects:
- dev-sandbox-1
- dev-sandbox-2
budgets:
monthly_limit: 1000
alert_thresholds: [50, 80, 100]
shared_vpc:
host_project: shared-vpc-host
service_projects:
- prod-app-1
- prod-app-2
- staging-app-1
Service Selection Guide
Compute Decision Matrix
| Requirement | Recommended | Alternatives |
|---|---|---|
| Stateless containers | Cloud Run | GKE Autopilot |
| Kubernetes workloads | GKE | GKE Autopilot, Cloud Run |
| Event-driven functions | Cloud Functions | Cloud Run |
| Long-running VMs | Compute Engine | GKE |
| Batch processing | Batch | Compute Engine, Cloud Run |
| ML training | Vertex AI | Compute Engine with GPUs |
Storage Decision Matrix
| Use Case | Service | Notes |
|---|---|---|
| Object storage | Cloud Storage | Standard/Nearline/Coldline/Archive |
| File storage | Filestore | NFSv3, high performance |
| VM disks | Persistent Disk | SSD/HDD, zonal/regional |
| Big data | Cloud Storage + BigQuery | Data lake + analytics |
| Block storage | Persistent Disk | Boot and data disks |
Database Decision Matrix
| Workload Type | Service | When to Use |
|---|---|---|
| Relational (PostgreSQL) | Cloud SQL | Traditional applications |
| Relational (MySQL) | Cloud SQL | Traditional applications |
| Relational (SQL Server) | Cloud SQL | Windows workloads |
| Global NoSQL | Firestore | Mobile, web apps |
| Wide-column | Bigtable | High throughput, low latency |
| Global relational | Spanner | Global consistency |
| Data warehouse | BigQuery | Analytics at scale |
| Cache | Memorystore | Redis/Memcached |
Security Best Practices
IAM Configuration
# gcp-iam-structure.yml
organization_policies:
constraints:
- constraint: constraints/iam.disableServiceAccountKeyCreation
enforce: true
- constraint: constraints/compute.disableNestedVirtualization
enforce: true
- constraint: constraints/compute.restrictSharedVpcSubnetworks
allowed_values:
- projects/shared-vpc-host/regions/*/subnetworks/*
service_accounts:
principle: one-service-account-per-application
naming: {app-name}@{project-id}.iam.gserviceaccount.com
key_management:
disable_key_creation: true
use_workload_identity: true
roles:
custom:
- name: app-deployer
permissions:
- run.services.get
- run.services.update
- artifactregistry.repositories.uploadArtifacts
predefined:
- roles/cloudsql.client
- roles/secretmanager.secretAccessor
- roles/pubsub.publisher
VPC Service Controls
# vpc-service-controls.yml
service_perimeter:
name: production-perimeter
resources:
- projects/prod-app-1
- projects/prod-app-2
restricted_services:
- bigquery.googleapis.com
- storage.googleapis.com
- sqladmin.googleapis.com
ingress_policies:
- from:
identities:
- serviceAccount:deployer@prod-app-1.iam.gserviceaccount.com
to:
resources: ["*"]
operations:
- bigquery.googleapis.com: ["bigquery.jobs.create"]
egress_policies:
- from:
identities:
- serviceAccount:exporter@prod-app-1.iam.gserviceaccount.com
to:
resources:
- projects/audit-logs
Network Security
# gcp-network-security.yml
firewall:
rules:
allow-internal:
source_ranges: [10.0.0.0/16]
allow: [tcp, udp, icmp]
allow-https:
source_ranges: [0.0.0.0/0]
allow: [tcp:443]
deny-all-egress:
direction: egress
deny: [all]
priority: 1000
cloud_armor:
security_policies:
- name: owasp-protection
rules:
- action: deny
priority: 1000
match:
expr: evaluatePreconfiguredExpr('sqli-stable')
- action: deny
priority: 1001
match:
expr: evaluatePreconfiguredExpr('xss-stable')
- action: allow
priority: 2147483647
match:
versioned_expr: SRC_IPS_V1
config:
src_ip_ranges: ["*"]
private_google_access:
enabled: true
subnetworks:
- app-subnet
- db-subnet
Cost Optimization
Committed Use Discounts
# gcp-cud-strategy.yml
committed_use_discounts:
compute:
vcpus: 100
memory: 400
term: 1-year
plan: general-purpose
resources:
- resource_type: compute
amount: 100
unit: vCPUs
- resource_type: memory
amount: 400
unit: GB
sustained_use_discounts:
automatic: true
applies_to:
- compute-engine
- gke-nodes
spot_vms:
use_for:
- batch-processing
- ci-cd-runners
- fault-tolerant-workloads
savings: up-to-91%
Storage Lifecycle
# gcp-storage-lifecycle.yml
lifecycle_policies:
application-logs:
rules:
- condition:
age: 30
action:
storage_class: nearline
- condition:
age: 90
action:
storage_class: coldline
- condition:
age: 365
action:
storage_class: archive
- condition:
age: 2555
action:
delete: true
access-logs:
rules:
- condition:
age: 7
action:
storage_class: nearline
- condition:
age: 30
action:
delete: true
Billing Management
# gcp-billing.yml
budgets:
monthly:
amount: 25000
currency: USD
alert_thresholds:
- 0.5
- 0.8
- 1.0
notifications:
- email: finance@company.com
- pubsub_topic: billing-alerts
billing_alerts:
anomaly_detection:
enabled: true
sensitivity: medium
quota_alerts:
enabled: true
thresholds:
- 80%
- 100%
labels:
required:
- environment
- cost-center
- application
- owner
Migration Patterns
Migrate for Compute Engine
# gcp-migration.yml
assessment:
tool: migrate-to-virtual-machines
discovery:
- vmware-vms
- aws-ec2
- azure-vms
fit_assessment:
- os_compatibility
- performance_requirements
- dependency_mapping
migration_phases:
wave_1:
workloads: test-non-critical
approach: lift-and-shift
timeline: 2_weeks
target: compute-engine
wave_2:
workloads: production
approach: lift-and-shift
timeline: 2_months
target: compute-engine
optimization:
approach: modernize
changes:
- migrate-to-cloud-run
- adopt-cloud-sql
- implement-cloud-storage
Anthos Hybrid Strategy
# anthos-hybrid.yml
strategy: hybrid-cloud
anthos:
on_prem:
platform: bare-metal
clusters:
- prod-on-prem
- dr-on-prem
gcp:
clusters:
- prod-gcp
- dr-gcp
config_management:
enabled: true
policy_controller:
enabled: true
constraints:
- require-pod-security-standards
- restrict-privileged-containers
service_mesh:
enabled: true
multi_cluster:
gateway: true