AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

gcp-cloud-architect

Use when designing or costing a Google Cloud architecture (Cloud Run, GKE, Functions, Cloud SQL, BigQuery, migration to GCP). Designs it with architecture_designer.py, deploys with deployment_manager.py and finds savings with cost_optimizer.py.

Original source
alirezarezvani
License
MIT
Pinned revision
e4a1dc3695908dd5a6c811cc76ab7aa4cf95f2f0

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

This skill also requires local files or dependencies. Remote reading alone may not be enough.

Ask your assistant

Use Agentic Memory skills to read alirezarezvani--gcp-cloud-architect at revision e4a1dc3695908dd5a6c811cc76ab7aa4cf95f2f0, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 4

references/best_practices.md

GCP Best Practices

Production-ready practices for naming, labels, IAM, networking, monitoring, and disaster recovery.


Table of Contents


Naming Conventions

Resource Naming Pattern

{environment}-{project}-{resource-type}-{purpose}

Examples:
  prod-myapp-gke-cluster
  dev-myapp-sql-primary
  staging-myapp-run-api
  prod-myapp-gcs-uploads

Project Naming

{org}-{team}-{environment}

Examples:
  acme-platform-prod
  acme-platform-dev
  acme-data-prod

Naming Rules

| Resource | Format | Max Length | Example | |----------|--------|-----------|---------| | Project ID | lowercase, hyphens | 30 chars | acme-platform-prod | | GKE Cluster | lowercase, hyphens | 40 chars | prod-api-cluster | | Cloud Run | lowercase, hyphens | 49 chars | prod-myapp-api | | Cloud SQL | lowercase, hyphens | 84 chars | prod-myapp-sql-primary | | GCS Bucket | lowercase, hyphens, dots | 63 chars | acme-prod-myapp-uploads | | Service Account | lowercase, hyphens | 30 chars | myapp-run-sa |


Labels and Organization

Required Labels

Apply these labels to all resources:

labels:
  environment: "prod"          # dev, staging, prod
  team: "platform"             # team owning the resource
  app: "myapp"                 # application name
  cost-center: "eng-001"       # billing allocation
  managed-by: "terraform"      # terraform, gcloud, console

Label-Based Cost Reporting

# Export billing data to BigQuery with labels
# Then query by label:
SELECT
  labels.value AS environment,
  SUM(cost) AS total_cost
FROM `billing_export.gcp_billing_export_v1_*`
CROSS JOIN UNNEST(labels) AS labels
WHERE labels.key = 'environment'
GROUP BY environment
ORDER BY total_cost DESC

Organization Hierarchy

Organization
├── Folder: Production
│   ├── Project: platform-prod
│   ├── Project: data-prod
│   └── Project: ml-prod
├── Folder: Non-Production
│   ├── Project: platform-dev
│   ├── Project: platform-staging
│   └── Project: data-dev
└── Folder: Shared Services
    ├── Project: shared-networking
    ├── Project: shared-security
    └── Project: shared-monitoring

IAM and Security

Principle of Least Privilege

# BAD: Basic roles are too broad
gcloud projects add-iam-policy-binding my-project \
  --member="user:[email protected]" \
  --role="roles/editor"

# GOOD: Use predefined roles
gcloud projects add-iam-policy-binding my-project \
  --member="user:[email protected]" \
  --role="roles/run.developer"

Service Account Best Practices

# 1. Create dedicated SA per workload
gcloud iam service-accounts create myapp-api-sa \
  --display-name="MyApp API Service Account"

# 2. Grant only required roles
gcloud projects add-iam-policy-binding my-project \
  --member="serviceAccount:[email protected]" \
  --role="roles/datastore.user"

# 3. Use Workload Identity for GKE (no key files)
gcloud iam service-accounts add-iam-policy-binding \
  [email protected] \
  --role="roles/iam.workloadIdentityUser" \
  --member="serviceAccount:my-project.svc.id.goog[default/myapp-api-ksa]"

# 4. NEVER download SA key files in production
# Instead, use attached service accounts or impersonation

VPC Service Controls

# Create a service perimeter to restrict data exfiltration
gcloud access-context-manager perimeters create my-perimeter \
  --title="Production Data Perimeter" \
  --resources="projects/123456" \
  --restricted-services="bigquery.googleapis.com,storage.googleapis.com" \
  --policy=$POLICY_ID

Organization Policies

# Restrict external IPs on VMs
gcloud resource-manager org-policies set-policy \
  --project=my-project policy.yaml

# policy.yaml
constraint: compute.vmExternalIpAccess
listPolicy:
  allValues: DENY

# Restrict public Cloud Storage
constraint: storage.publicAccessPrevention
booleanPolicy:
  enforced: true

Encryption

| Layer | Service | Default | |-------|---------|---------| | At rest | Google-managed keys | Always enabled | | At rest | CMEK (Cloud KMS) | Optional, recommended | | In transit | TLS 1.3 | Always enabled | | Application | Cloud KMS | Encrypt sensitive fields |

# Create CMEK key for Cloud SQL
gcloud kms keys create myapp-sql-key \
  --keyring=myapp-keyring \
  --location=us-central1 \
  --purpose=encryption

# Use CMEK with Cloud SQL
gcloud sql instances create myapp-db \
  --disk-encryption-key=projects/my-project/locations/us-central1/keyRings/myapp-keyring/cryptoKeys/myapp-sql-key

Networking

VPC Design

# Create custom VPC (avoid default network)
gcloud compute networks create myapp-vpc \
  --subnet-mode=custom

# Create subnets with secondary ranges for GKE
gcloud compute networks subnets create myapp-subnet \
  --network=myapp-vpc \
  --region=us-central1 \
  --range=10.0.0.0/20 \
  --secondary-range pods=10.4.0.0/14,services=10.8.0.0/20 \
  --enable-private-google-access

Shared VPC

Use Shared VPC for multi-project environments:

Host Project (shared-networking)
├── VPC: shared-vpc
│   ├── Subnet: prod-us-central1 → Service Project: platform-prod
│   ├── Subnet: prod-europe-west1 → Service Project: platform-prod
│   └── Subnet: dev-us-central1 → Service Project: platform-dev

Firewall Rules

# Allow internal traffic
gcloud compute firewall-rules create allow-internal \
  --network=myapp-vpc \
  --allow=tcp,udp,icmp \
  --source-ranges=10.0.0.0/8

# Allow health checks from Google load balancers
gcloud compute firewall-rules create allow-health-checks \
  --network=myapp-vpc \
  --allow=tcp:8080 \
  --source-ranges=35.191.0.0/16,130.211.0.0/22 \
  --target-tags=allow-health-check

# Deny all other ingress (implicit, but be explicit)
gcloud compute firewall-rules create deny-all-ingress \
  --network=myapp-vpc \
  --action=DENY \
  --rules=all \
  --direction=INGRESS \
  --priority=65534

Private Google Access

Always enable Private Google Access to reach GCP APIs without public IPs:

gcloud compute networks subnets update myapp-subnet \
  --region=us-central1 \
  --enable-private-google-access

Monitoring and Logging

Cloud Monitoring Setup

# Create uptime check
gcloud monitoring uptime create \
  --display-name="API Health Check" \
  --resource-type=cloud-run-revision \
  --resource-labels="service_name=myapp-api,location=us-central1" \
  --check-request-path="/health" \
  --period=60s

# Create alerting policy
gcloud alpha monitoring policies create \
  --display-name="High Error Rate" \
  --condition-display-name="Cloud Run 5xx > 1%" \
  --condition-filter='resource.type="cloud_run_revision" AND metric.type="run.googleapis.com/request_count" AND metric.labels.response_code_class="5xx"' \
  --condition-threshold-value=1 \
  --notification-channels="projects/my-project/notificationChannels/12345"

Key Metrics to Monitor

| Service | Metric | Alert Threshold | |---------|--------|-----------------| | Cloud Run | request_latencies (p99) | >2s | | Cloud Run | request_count (5xx) | >1% of total | | Cloud SQL | cpu/utilization | >80% | | Cloud SQL | disk/utilization | >85% | | GKE | container/cpu/utilization | >80% | | GKE | node/cpu/allocatable_utilization | >85% | | Pub/Sub | subscription/oldest_unacked_message_age | >300s | | BigQuery | query/execution_time | >60s |

Log-Based Metrics

# Create a metric for application errors
gcloud logging metrics create app-errors \
  --description="Application error count" \
  --log-filter='resource.type="cloud_run_revision" AND severity>=ERROR'

# Create log sink to BigQuery for analysis
gcloud logging sinks create audit-logs-bq \
  bigquery.googleapis.com/projects/my-project/datasets/audit_logs \
  --log-filter='logName="projects/my-project/logs/cloudaudit.googleapis.com%2Factivity"'

Log Exclusion (Cost Reduction)

# Exclude verbose debug logs to save on Cloud Logging costs
gcloud logging sinks create _Default \
  --log-filter='NOT (severity="DEBUG" OR severity="DEFAULT")' \
  --description="Exclude debug-level logs"

# Or create exclusion filters
gcloud logging exclusions create exclude-debug \
  --log-filter='severity="DEBUG"' \
  --description="Exclude debug logs to reduce costs"

Cost Optimization

Committed Use Discounts

| Term | Compute Discount | Memory Discount | |------|-----------------|-----------------| | 1 year | 37% | 37% | | 3 years | 55% | 55% |

# Check recommendations
gcloud recommender recommendations list \
  --project=my-project \
  --location=us-central1 \
  --recommender=google.compute.commitment.UsageCommitmentRecommender

Sustained Use Discounts

Automatic discounts for resources running >25% of the month:

| Usage | Discount | |-------|----------| | 25-50% | 20% | | 50-75% | 40% | | 75-100% | 60% |

BigQuery Cost Control

-- Use partitioning to limit data scanned
CREATE TABLE my_dataset.events
PARTITION BY DATE(timestamp)
CLUSTER BY event_type
AS SELECT * FROM raw_events;

-- Estimate query cost before running
-- Use --dry_run flag
bq query --dry_run --use_legacy_sql=false \
  'SELECT * FROM my_dataset.events WHERE DATE(timestamp) = "2026-01-01"'

Cloud Storage Optimization

# Enable Autoclass for automatic class management
gsutil mb -l us-central1 --autoclass gs://my-bucket/

# Set lifecycle policy
gsutil lifecycle set lifecycle.json gs://my-bucket/

Disaster Recovery

RPO/RTO Targets

| Tier | RPO | RTO | Strategy | |------|-----|-----|----------| | Tier 1 (Critical) | 0 | <1 hour | Multi-region active-active | | Tier 2 (Important) | <1 hour | <4 hours | Regional HA + cross-region backup | | Tier 3 (Standard) | <24 hours | <24 hours | Automated backups + restore |

Backup Strategy

# Cloud SQL automated backups
gcloud sql instances patch myapp-db \
  --backup-start-time=02:00 \
  --enable-point-in-time-recovery

# Firestore scheduled exports
gcloud firestore export gs://myapp-backups/firestore/$(date +%Y%m%d)

# GKE cluster backup with Backup for GKE
gcloud beta container backup-restore backup-plans create myapp-plan \
  --project=my-project \
  --location=us-central1 \
  --cluster=projects/my-project/locations/us-central1/clusters/myapp-cluster \
  --all-namespaces \
  --cron-schedule="0 2 * * *"

Multi-Region Failover

# Cloud SQL cross-region replica for DR
gcloud sql instances create myapp-db-replica \
  --master-instance-name=myapp-db \
  --region=us-east1

# Promote replica during failover
gcloud sql instances promote-replica myapp-db-replica

Common Pitfalls

Technical Debt

| Pitfall | Solution | |---------|----------| | Using default VPC | Always create custom VPCs | | Not enabling audit logs | Enable Cloud Audit Logs from day one | | Single-region deployment | Plan for multi-zone at minimum | | No IaC | Use Terraform from the start |

Security Mistakes

| Mistake | Prevention | |---------|------------| | SA key files in code | Use Workload Identity, attached SAs | | Public GCS buckets | Enable org policy for public access prevention | | Basic roles (Owner/Editor) | Use predefined or custom roles | | No encryption key management | Use CMEK for sensitive data | | Default service account | Create dedicated SAs per workload |

Performance Issues

| Issue | Solution | |-------|----------| | Cold starts on Cloud Run | Set min-instances=1 for latency-critical services | | Slow BigQuery queries | Partition tables, use clustering, avoid SELECT * | | GKE pod scheduling delays | Use PodDisruptionBudget, pre-provision with Autopilot | | Firestore hotspots | Distribute writes across document IDs evenly |

Cost Surprises

| Surprise | Prevention | |----------|------------| | Undeleted resources | Label everything, review weekly | | Egress costs | Keep traffic in same region, use Private Google Access | | Cloud NAT charges | Use Private Google Access for GCP service traffic | | Log ingestion costs | Set exclusion filters for debug/verbose logs | | BigQuery full scans | Always use partitioning and clustering | | Idle GKE clusters | Delete dev clusters nightly, use Autopilot |