[Feat]: Add environment variable to disable A2A SDK telemetry when OpenTelemetry is installed by another library #613
Description
Is your feature request related to a problem? Please describe.
When using a2a-sdk[http-server] alongside observability tools like Langfuse that bring OpenTelemetry as a dependency, the A2A SDK's @trace_class decorators automatically activate and emit spans for internal operations (EventQueue, InMemoryTaskStore, etc.).
This pollutes trace dashboards with unwanted internal spans like:
a2a.server.events.event_queue
a2a.server.events.in_memory_task_store
The current _NoOp fallback in telemetry.py only works when OpenTelemetry isn't installed at all — not when it's present but unwanted for A2A internals.
Describe the solution you'd like
Add an environment variable (e.g., A2A_TELEMETRY_DISABLED=true) that forces the _NoOp path in telemetry.py regardless of whether OpenTelemetry is installed.
pythonimport os
if os.getenv("A2A_TELEMETRY_DISABLED", "").lower() == "true":
trace = _NoOp()
else:
try:
from opentelemetry import trace
...
Describe alternatives you've considered
Custom OTEL Sampler: Filter spans by name — adds complexity
Monkey-patching a2a.utils.telemetry: Fragile, breaks on SDK updates
Uninstalling OpenTelemetry: Not possible when other tools require it
Additional context
Similar pattern exists in other SDKs (e.g., Vercel AI SDK's experimental_telemetry option). This would allow users to opt-out of A2A tracing while still using OpenTelemetry for their own instrumentation
Code of Conduct
- I agree to follow this project's Code of Conduct