Roger Y.

os.environ["OTEL_PYTHON_TRACER_PROVIDER"] = "otel_sdk_tracer_provider"

from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk import trace as trace_sdk
from opentelemetry.sdk.resources import Resource
from opentelemetry.sdk.trace.export import SimpleSpanProcessor

    resource = Resource(attributes={"openinference.project.name": "default"})
    tracer_provider = trace_sdk.TracerProvider(resource=resource)

    endpoint = "https://app.phoenix.arize.com/s/elena-kosobrodova/v1/traces"
    headers = {"authorization": f"Bearer {phoenix_api_key}"}

    exporter = OTLPSpanExporter(endpoint=endpoint, headers=headers)
    tracer_provider.add_span_processor(SimpleSpanProcessor(exporter))

    GoogleADKInstrumentor().instrument(tracer_provider=tracer_provider)

If you deploy your app using --entrypoint-object=root_agent (or app), Vertex AI serves the ADK agent directly. Because the ADK Agent class doesn't use your custom set_up() method, Vertex AI silently skips it during container initialization. Your agent responds to queries perfectly, but the OpenTelemetry provider is never initialized.