cloud-run-deployer

cloud-run-deployer

Containerize and deploy an ADK 2.0 agent to Cloud Run.

Project layout

agent_service/
├── agent.py            # has root_agent
├── server.py           # FastAPI wrapper
├── requirements.txt
├── Dockerfile
└── .gcloudignore

Dockerfile

FROM python:3.12-slim

WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

COPY . .

ENV PORT=8080
ENV PYTHONUNBUFFERED=1

CMD exec uvicorn server:app --host 0.0.0.0 --port $PORT

requirements.txt

google-adk[extensions] >= 2.0.0
fastapi
uvicorn[standard]

server.py — FastAPI wrapper

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from google.adk.runners import Runner
from google.adk.sessions import VertexAiSessionService
from agent import root_agent
import os

app = FastAPI()

session_service = VertexAiSessionService(
    project=os.environ["GOOGLE_CLOUD_PROJECT"],
    location=os.environ["GOOGLE_CLOUD_LOCATION"],
)
runner = Runner(agent=root_agent, session_service=session_service)


class ChatRequest(BaseModel):
    user_id: str
    session_id: str | None = None
    message: str


@app.post("/chat")
async def chat(req: ChatRequest):
    if req.session_id:
        session = await session_service.get_session(req.session_id)
    else:
        session = await session_service.create_session(
            app_name="my_agent", user_id=req.user_id
        )
    response = await runner.run_async(session=session, input=req.message)
    return {"session_id": session.id, "response": response.text}


@app.get("/health")
def health():
    return {"status": "ok"}

Deploy

PROJECT=my-project
REGION=us-central1
SERVICE=my-adk-agent

gcloud run deploy $SERVICE \
    --source . \
    --region $REGION \
    --platform managed \
    --allow-unauthenticated \
    --set-env-vars "GOOGLE_GENAI_USE_VERTEXAI=true,GOOGLE_CLOUD_PROJECT=$PROJECT,GOOGLE_CLOUD_LOCATION=$REGION" \
    --memory 1Gi \
    --cpu 1 \
    --min-instances 0 \
    --max-instances 10 \
    --concurrency 20

For private services, drop --allow-unauthenticated and use IAM-bound invokers.

Vertex auth (recommended for prod)

Cloud Run automatically uses the runtime service account. Grant it:

• roles/aiplatform.user (Vertex AI / Gemini)
• roles/aiplatform.sessionUser (Vertex sessions)

SA=$(gcloud run services describe $SERVICE --region $REGION --format='value(spec.template.spec.serviceAccountName)')
gcloud projects add-iam-policy-binding $PROJECT --member="serviceAccount:$SA" --role="roles/aiplatform.user"

.gcloudignore

__pycache__/
.pytest_cache/
*.pyc
.env
.venv/
tests/

Validation

• curl https://<service>/health returns {"status": "ok"}
• POST to /chat returns a valid response
• Cloud logging shows model/tool events
• Cold-start time acceptable (~2-5s; tune --min-instances 1 for hot)

See also

• gke-deployer for higher-control Kubernetes deployment
• vertex-agent-engine-deployer for managed agent runtime