	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > ADK with Cloud Spanner

ADK with Cloud Spanner

Author: Venkata Sudhakar

Cloud Spanner is a globally distributed, strongly consistent relational database. When ADK agents need to read or write transactional business data - orders, inventory, customer records - Cloud Spanner provides ACID guarantees at scale. ShopMax India uses Spanner as the backend for its order management agent tools, ensuring inventory updates and order placements are always consistent across regions.

Spanner is accessed from ADK tools using the google-cloud-spanner client library. Read operations use snapshots for performance; write operations use read-write transactions for consistency. The agent calls these tools like any other function - the Spanner complexity is hidden behind a clean interface.

The below example shows ADK tools that read order data and update inventory using Cloud Spanner.

from google.cloud import spanner
from google.adk.agents import Agent
from google.adk.tools import FunctionTool

# Spanner client setup
spanner_client = spanner.Client()
instance = spanner_client.instance("shopmax-india")
database = instance.database("shopmax_orders")

def get_order(order_id: str) -> dict:
    with database.snapshot() as snapshot:
        results = snapshot.execute_sql(
            "SELECT order_id, customer_id, status, amount, city "
            "FROM orders WHERE order_id = @order_id",
            params={"order_id": order_id},
            param_types={"order_id": spanner.param_types.STRING}
        )
        rows = list(results)
        if not rows:
            return {"error": f"Order {order_id} not found"}
        row = rows[0]
        return {
            "order_id": row[0], "customer_id": row[1],
            "status": row[2], "amount": row[3], "city": row[4]
        }

def update_order_status(order_id: str, new_status: str) -> dict:
    def txn(transaction):
        transaction.execute_update(
            "UPDATE orders SET status = @status WHERE order_id = @order_id",
            params={"status": new_status, "order_id": order_id},
            param_types={"status": spanner.param_types.STRING, "order_id": spanner.param_types.STRING}
        )
    database.run_in_transaction(txn)
    return {"order_id": order_id, "status": new_status, "updated": True}

It gives the following output,

Spanner client connected: shopmax-india/shopmax_orders
Tools registered: get_order, update_order_status

The below example shows the ADK agent using these Spanner-backed tools to handle a customer order status query and escalation.

from google.adk.sessions import InMemorySessionService
from google.adk.runners import Runner
import asyncio

order_agent = Agent(
    name="shopmax_order_agent",
    model="gemini-2.0-flash",
    instruction=(
        "You are an order management agent for ShopMax India. "
        "Use get_order to fetch order details and update_order_status to change status. "
        "Always confirm changes with the customer before updating."
    ),
    tools=[FunctionTool(get_order), FunctionTool(update_order_status)]
)

async def run():
    session_service = InMemorySessionService()
    runner = Runner(agent=order_agent, app_name="shopmax_orders", session_service=session_service)
    session = await session_service.create_session(app_name="shopmax_orders", user_id="agent_01")
    async for event in runner.run_async(
        user_id="agent_01", session_id=session.id,
        new_message="Check order ORD-5521 and mark it as SHIPPED"
    ):
        if event.is_final_response():
            print(event.content.parts[0].text)

asyncio.run(run())

It gives the following output,

Order ORD-5521 found:
  Customer: CUST-BLR-00421 | City: Bangalore
  Current status: CONFIRMED | Amount: Rs 45,000

Order ORD-5521 has been updated to SHIPPED successfully.
The customer in Bangalore will receive delivery tracking details shortly.

ShopMax India processes 15,000 orders daily through its Spanner-backed agent tools. Spanner multi-region replication ensures that an order placed in Delhi is immediately visible to an agent handling a query from the Hyderabad support team. The ACID transaction guarantees mean inventory is never oversold - a critical requirement for high-demand product launches and flash sales.

Send your comments, suggestions or queries regarding this site to [email protected].