Help Needed with Phoenix Integration for Google ADK Agent on AgentEngine Deployment | Arize AI Community

Arize AI Community Icon

Help Needed with Phoenix Integration for Google ADK Agent on AgentEngine Deployment | Arize AI Community

RunLLM
APP
·
·
For Google AgentEngine deployments, instrumentation must be configured inside the remote agent module, not just in your main application. You need to ensure that the Phoenix tracing setup (including the `register` function and the relevant instrumentor, e.g., `GoogleADKInstrumentor().instrument(tracer_provider=tracer_provider)`) is included in your agent module code that runs on AgentEngine. Also, make sure all required packages are listed in your deployment requirements and that environment variables for the Phoenix endpoint and API key are set in the remote environment. Setting `GOOGLE_CLOUD_AGENT_ENGINE_ENABLE_TELEMETRY=False` is correct, as Phoenix uses OpenTelemetry directly, not the built-in telemetry. See the official <https://arize.com/docs/phoenix/integrations/python/google-adk/google-adk-tracing#agent-engine-deployment|Phoenix Google ADK Agent Engine deployment guide> for details. If traces appear locally but not in AgentEngine, it's likely the remote agent module is missing the tracing setup or the environment variables are not correctly passed to the remote environment. Double-check that the tracing code is executed in the remote agent context and that the Phoenix endpoint and API key are available there. Would you like a step-by-step deployment example? Sources: - <https://arize.com/docs/phoenix/integrations/python/google-adk/google-adk-tracing#agent-engine-deployment|Phoenix Google ADK Agent Engine deployment guide>
💡 Hint: Mention RunLLM in the thread for followups.
Elena k.
·
·
GOOGLE_CLOUD_AGENT_ENGINE_ENABLE_TELEMETRY=False and I use GoogleADKInstrumentor().instrument(tracer_provider=tracer_provider)
Elena k.
·
·
then in agent.py:
Elena k.
·
·
the Phoenix endpoint and API key are available and correct, because I see the traces when I run the agent locally
Roger Y.
·
·
Hi Elena. Was the Phoenix Google ADK Agent Engine deployment guide (referenced above by the bot) helpful for you?
Elena k.
·
·
Hey Roger. No, it didn't help. As I said, I've already used GOOGLE_CLOUD_AGENT_ENGINE_ENABLE_TELEMETRY=False and GoogleADKInstrumentor().instrument(tracer_provider=tracer_provider)
Roger Y.
·
·
just to confirm, you tried this step and it had no effect?
Elena k.
·
·
I didn't set OTEL_EXPORTER_OTLP_ENDPOINT. but I used this:
Elena k.
·
·
requirements.txt:
Elena k.
·
·
I think, arize-otel is for Arize, not for Phoenix
Roger Y.
·
·
it should be exchangeable. let me ask the team
Elena k.
·
·
thank you
Roger Y.
·
·
what does your deploy script look like?

·

·

something like this?

import os
import vertexai
from vertexai import agent_engines

project_id = os.environ.get("GOOGLE_CLOUD_PROJECT")
if not project_id:
    raise RuntimeError("Set GOOGLE_CLOUD_PROJECT before deploying.")

# Initialize the Vertex AI client
vertexai.init(project=project_id, location=os.environ.get("GOOGLE_CLOUD_LOCATION", "us-central1"))
client = vertexai.Client(project=project_id, location=os.environ.get("GOOGLE_CLOUD_LOCATION", "us-central1"))

print("Deploying ADK Agent to Vertex AI Agent Engine...")

remote_agent = client.agent_engines.create(
    # Tell Agent Engine to run the 'agent_engine' object inside 'agent.py'
    agent_engine=agent_engines.ModuleAgent(module_name="agent", agent_name="agent_engine"),
    config={
        "requirements": [
            "google-cloud-aiplatform[agent_engines,adk]>=1.139.0",
            "openinference-instrumentation-google-adk>=0.1.9",
            "arize-phoenix-otel>=0.15.0",
        ],
        "extra_packages": [
            "agent.py",
            "agent_engine_app.py",
        ],
        "env_vars": {
            "GOOGLE_CLOUD_PROJECT": project_id,
            "PHOENIX_COLLECTOR_ENDPOINT": os.environ.get("PHOENIX_COLLECTOR_ENDPOINT", "https://app.phoenix.arize.com/v1/traces"),
            "PHOENIX_PROJECT_NAME": os.environ.get("PHOENIX_PROJECT_NAME", "default"),
            "PHOENIX_API_KEY": os.environ.get("PHOENIX_API_KEY", ""),
        }
    }
)

print(f"Deployment complete! Agent Resource Name: {remote_agent.api_resource.name}")

Elena k.
·
·
it works fine. The agent is deployed to AgentEngine and the agent itself works. Environment variables: OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=True, GOOGLE_GENAI_USE_VERTEXAI=True, GOOGLE_CLOUD_AGENT_ENGINE_ENABLE_TELEMETRY=False