GCP Patterns

SkillDatabases & data

Cloud Run deployment, BigQuery optimization, Pub/Sub patterns, IAM best practices

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the GCP Patterns skill

What this skill tells your AI

The instructions your AI receives, as published by vibeeval/vibecosystem in skills/gcp-patterns/SKILL.md and read by ahel’s review.

Cloud Run Deployment

Dockerfile for Cloud Run

FROM node:20-slim AS builder
WORKDIR /app
COPY package*.json ./
RUN npm ci --production=false
COPY . .
RUN npm run build

FROM node:20-slim
WORKDIR /app
RUN addgroup --system app && adduser --system --ingroup app app
COPY --from=builder /app/dist ./dist
COPY --from=builder /app/node_modules ./node_modules
COPY --from=builder /app/package.json ./
USER app
EXPOSE 8080
ENV PORT=8080 NODE_ENV=production
CMD ["node", "dist/server.js"]

Cloud Run Service YAML

apiVersion: serving.knative.dev/v1
kind: Service
metadata:
  name: order-service
  annotations:
    run.googleapis.com/launch-stage: GA
spec:
  template:
    metadata:
      annotations:
        autoscaling.knative.dev/minScale: "1"
        autoscaling.knative.dev/maxScale: "100"
        run.googleapis.com/cpu-throttling: "false"
        run.googleapis.com/startup-cpu-boost: "true"
    spec:
      containerConcurrency: 80
      timeoutSeconds: 300
      serviceAccountName: order-service@project-id.iam.gserviceaccount.com
      containers:
        - image: gcr.io/project-id/order-service:latest
          ports:
            - containerPort: 8080
          resources:
            limits:
              cpu: "2"
              memory: 1Gi
          env:
            - name: DB_CONNECTION
              valueFrom:
                secretKeyRef:
                  key: latest
                  name: db-connection-string
          startupProbe:
            httpGet:
              path: /healthz
              port: 8080
            initialDelaySeconds: 5
            periodSeconds: 3

Deploy Command

gcloud run deploy order-service \
  --image gcr.io/$PROJECT_ID/order-service:$GIT_SHA \
  --region us-central1 \
  --service-account order-service@$PROJECT_ID.iam.gserviceaccount.com \
  --set-secrets "DB_URL=db-connection:latest" \
  --min-instances 1 \
  --max-instances 100 \
  --cpu 2 --memory 1Gi \
  --concurrency 80 \
  --no-allow-unauthenticated

BigQuery Optimization

-- Use partitioning and clustering
CREATE TABLE `project.dataset.events`
PARTITION BY DATE(event_timestamp)
CLUSTER BY user_id, event_type
AS SELECT * FROM `project.dataset.raw_events`;

-- Always filter on partition column
SELECT event_type, COUNT(*) as cnt
FROM `project.dataset.events`
WHERE event_timestamp BETWEEN '2025-01-01' AND '2025-01-31'
  AND event_type = 'purchase'
GROUP BY event_type;

-- Use approximate functions for large datasets
SELECT APPROX_COUNT_DISTINCT(user_id) as unique_users
FROM `project.dataset.events`
WHERE DATE(event_timestamp) = CURRENT_DATE();

-- Avoid SELECT * (scans all columns, costs more)
-- Use column selection and LIMIT for exploration

Pub/Sub Patterns

from google.cloud import pubsub_v1
from google.api_core import retry
import json

# Publisher with ordering and retry
publisher = pubsub_v1.PublisherClient()
topic_path = publisher.topic_path("project-id", "order-events")

def publish_event(event: dict, ordering_key: str = "") -> str:
    data = json.dumps(event).encode("utf-8")
    future = publisher.publish(
        topic_path,
        data,
        ordering_key=ordering_key,
        event_type=event["type"],
    )
    return future.result(timeout=30)

# Subscriber with exactly-once processing
subscriber = pubsub_v1.SubscriberClient()
subscription_path = subscriber.subscription_path("project-id", "order-events-sub")

def callback(message: pubsub_v1.types.PubsubMessage) -> None:
    try:
        event = json.loads(message.data.decode("utf-8"))
        idempotency_key = message.message_id

        if already_processed(idempotency_key):
            message.ack()
            return

        process_event(event)
        mark_processed(idempotency_key)
        message.ack()
    except Exception as e:
        logger.error(f"Failed to process message: {e}")
        message.nack()

subscriber.subscribe(subscription_path, callback=callback)

IAM Best Practices

Principles:
  - Least privilege: grant minimum permissions needed
  - Service accounts per service (not shared)
  - No user accounts in production workloads
  - Prefer predefined roles over primitive roles

Per-Service Pattern:
  order-service:
    roles:
      - roles/cloudsql.client          # DB access
      - roles/pubsub.publisher         # Publish events
      - roles/secretmanager.secretAccessor  # Read secrets
    # NOT: roles/editor (too broad)

Workload Identity (GKE):
  - Bind K8s SA to GCP SA
  - No key files, automatic credential rotation

Checklist

  • Cloud Run services use dedicated service accounts
  • Secrets stored in Secret Manager, not env vars
  • BigQuery tables partitioned and clustered
  • Pub/Sub subscribers implement idempotent processing
  • Health check endpoints configured for all services
  • Min instances set for latency-sensitive services
  • IAM follows least privilege (no primitive roles)
  • Cloud Armor WAF in front of public endpoints
  • VPC connector for private resource access

Anti-Patterns

  • Using default compute service account (overprivileged)
  • SELECT * on BigQuery (scans all columns, high cost)
  • Pub/Sub without dead letter queue (messages lost on repeated failure)
  • Hardcoding project ID instead of using environment detection
  • Not setting concurrency limits on Cloud Run (OOM under load)
  • Using Cloud Run for long-running background jobs (use Cloud Tasks)
  • Storing secrets in environment variables instead of Secret Manager

Signals

GitHub stars
531
Forks
44
Last commit
Aug 2026
Advanced
Catalog kind
skill
Gateway key
gcp-patterns
Source
github.com/vibeeval/vibecosystem