Managing Cloud Run, AI Agents and Developer Environments

What it is

This page covers operating serverless and AI workloads: deploying new Cloud Run revisions, splitting traffic between them, configuring Cloud Run autoscaling, deploying an agent to Agent Runtime on Gemini Enterprise Agent Platform, managing notebooks in Agent Platform Workbench and BigQuery, and managing developer environments with Cloud Workstations.

How it works

New revisions and traffic splitting on Cloud Run

Each deploy creates a new revision. By default it receives all traffic, but you can deploy it with none and shift gradually.

# deploy a new revision without sending it traffic, tagged for testing
gcloud run deploy orders --image=us-docker.pkg.dev/PROJECT/repo/orders:2.0 \
  --region=us-central1 --no-traffic --tag=v2

# send 10 percent of traffic to the new revision
gcloud run services update-traffic orders --region=us-central1 \
  --to-revisions=orders-00002-abc=10

# send all traffic to the latest revision
gcloud run services update-traffic orders --region=us-central1 --to-latest

# roll back by sending traffic to the previous revision
gcloud run services update-traffic orders --region=us-central1 \
  --to-revisions=orders-00001-xyz=100

A tag gives the revision its own URL for testing before it gets production traffic. Cloud Run functions also run as Cloud Run services, so their revisions can be split the same way.

Cloud Run autoscaling

gcloud run services update orders --region=us-central1 --min-instances=1 --max-instances=50 --concurrency=80

Deploying an agent to Agent Runtime

Agent Runtime on Gemini Enterprise Agent Platform (formerly Vertex AI Agent Engine) is a managed service to deploy, scale and manage AI agents in production. Developers build the agent with a supported framework (such as Google's Agent Development Kit), then deploy it with the Agent Platform SDK; Agent Runtime handles infrastructure, scaling, sessions and integration with monitoring and IAM.

Notebooks

Managing Workbench instances includes starting and stopping them (to save cost), choosing machine types and GPUs, enabling idle shutdown, and controlling access with IAM.

Cloud Workstations

Cloud Workstations provides managed, secure development environments in Google Cloud. Administrators define workstation configurations (machine type, container image with the IDE and tools, network, idle timeouts); developers start workstations from them in the browser or a local IDE. Source code stays in the cloud, environments are consistent, and access is controlled by IAM.

A worked scenario

A team releases version 2 of its Cloud Run API with --no-traffic --tag=v2, tests the tagged URL, moves 10 percent of traffic, watches error rates, and then sends 100 percent to the latest revision. They set min instances to 1 to remove cold starts. Their data scientists use Workbench instances with idle shutdown, analysts use BigQuery notebooks, and contractors get Cloud Workstations so code never leaves the cloud. A support agent is deployed to Agent Runtime.

How the exam asks about this

• "Deploy a new version without serving traffic to it." gcloud run deploy --no-traffic.
• "Send 20 percent of traffic to the new revision." gcloud run services update-traffic --to-revisions=REV=20.
• "Avoid cold starts." Set minimum instances.
• "Managed, secure developer environments with source code kept in the cloud." Cloud Workstations.
• "Deploy and scale an AI agent with a managed service." Agent Runtime.