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

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > ADK Distributed Tracing with Cloud Trace

ADK Distributed Tracing with Cloud Trace

Author: Venkata Sudhakar

Distributed tracing gives visibility into exactly where time is spent during an ADK agent request - how long the model took to respond, which tools were called, how long each tool took, and where errors occurred. ShopMax India uses Cloud Trace with OpenTelemetry to build a complete picture of every agent interaction, from the HTTP request entry point through model calls and downstream tool invocations.

OpenTelemetry (OTel) is the industry standard for instrumentation. The opentelemetry-sdk-python package integrates with Cloud Trace via the google-cloud-trace exporter. Each logical unit of work becomes a span - the agent handler creates a root span, and each tool call creates a child span. Trace IDs are propagated in HTTP headers so spans across services are linked into a single trace in the Cloud Trace console.

The below example shows an ADK agent with full OpenTelemetry tracing instrumented across the request handler and tool calls.

from opentelemetry import trace
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor
from opentelemetry.exporter.cloud_trace import CloudTraceSpanExporter
from opentelemetry.sdk.resources import Resource
from google.adk.agents import LlmAgent
from google.adk.tools import FunctionTool
from google.adk.sessions import InMemorySessionService
from google.adk.runners import Runner
import time, os

# Configure Cloud Trace exporter
resource = Resource.create({"service.name": "shopmax-support-agent"})
provider = TracerProvider(resource=resource)
provider.add_span_processor(
    BatchSpanProcessor(CloudTraceSpanExporter(project_id=os.environ["GOOGLE_CLOUD_PROJECT"]))
)
trace.set_tracer_provider(provider)
tracer = trace.get_tracer("shopmax.agent")

def check_order_status(order_id: str) -> dict:
    """Check the status of a ShopMax order."""
    with tracer.start_as_current_span("tool.check_order_status") as span:
        span.set_attribute("order.id", order_id)
        time.sleep(0.05)  # simulate DB lookup
        result = {"order_id": order_id, "status": "shipped", "eta": "2026-04-08"}
        span.set_attribute("order.status", result["status"])
        return result

agent = LlmAgent(
    name="shopmax_support_agent",
    model="gemini-2.0-flash",
    instruction="You are a ShopMax India support agent. Help customers track orders.",
    tools=[FunctionTool(check_order_status)],
)

session_service = InMemorySessionService()
runner = Runner(agent=agent, session_service=session_service)

with tracer.start_as_current_span("agent.request") as root_span:
    root_span.set_attribute("user.id", "user_mumbai_001")
    root_span.set_attribute("session.id", "sess_trace_001")
    events = list(runner.run(
        user_id="user_mumbai_001",
        session_id="sess_trace_001",
        new_message={"role": "user", "parts": [{"text": "Where is my order ORD-5042?"}]},
    ))
    reply = next((e.content.parts[0].text for e in events if e.is_final_response()), "")
    root_span.set_attribute("agent.reply_length", len(reply))
    print("Reply:", reply)

provider.force_flush()
print("Trace exported to Cloud Trace.")

It gives the following output,

Reply: Your order ORD-5042 has been shipped and is expected to arrive by 8 April 2026.
Trace exported to Cloud Trace.

# Cloud Trace console shows:
# agent.request                          1,842 ms (root span)
#   tool.check_order_status                 52 ms
#   gemini.generate_content              1,780 ms

In production, propagate the W3C trace context header (traceparent) from the incoming HTTP request into the root span so that traces link across services - the API gateway, the ADK agent, and any downstream microservices all appear as a single trace. Create Cloud Trace latency distributions in Cloud Monitoring to alert when the p99 agent response time exceeds 5 seconds, giving the ShopMax on-call team early warning before customers notice degraded performance.

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