Failed to detach context/Token has already been used once errors with instrument_anthropic()

[francisjervis]

Description

I am getting intermittent (apparently non-deterministic) opentelemetry errors when instrumenting the Anthropic client with Logfire inside a Flask app. This may well be the same issue discussed here: open-telemetry/opentelemetry-python#3919 based on the error logs posted there.

ERROR:opentelemetry.context:Failed to detach context
Traceback (most recent call last):
  File "./.venv/lib/python3.11/site-packages/flask/helpers.py", line 132, in generator
    yield from gen
GeneratorExit

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "./.venv/lib/python3.11/site-packages/opentelemetry/context/__init__.py", line 155, in detach
    _RUNTIME_CONTEXT.detach(token)
  File "..venv/lib/python3.11/site-packages/opentelemetry/context/contextvars_context.py", line 53, in detach
    self._current_context.reset(token)
RuntimeError: <Token used var=<ContextVar name='current_context' default={} at 0x10cbb99e0> at 0x163a1ef40> has already been used once

Python, Logfire & OS Versions, related packages (not required)

Python 3.11, Mac OS 26.2 (25C56)

Logfire 4.17

[alexmojaki]

Thanks. I've commented on the OTEL issue. Does this happen in tests or an actual app? Does the anthropic instrumentation actually have anything to do with it?

[alexmojaki]

Also just FYI this log is annoying but I don't think it should be affecting anything or a sign that you need to be concerned.

[francisjervis]

Thanks for looking at this. It's happening in a multi-generation workflow in an actual app. It does in fact appear to cause a crash when the streaming API is being used - specifically, when the async Anthropic client is being used the error breaks the generator, as indicated in the logs posted.

[alexmojaki]

Added another comment there. The problem appears to about flask streaming. I still doubt it relates to Anthropic.

The log comes from this code in OTel:

    try:
        _RUNTIME_CONTEXT.detach(token)
    except Exception:  # pylint: disable=broad-exception-caught
        logger.exception("Failed to detach context")

specifically the logger.exception. The exception is caught. I'd be really surprised if anything is actually crashing.

[francisjervis]

Something is definitely crashing but it would be fair to say I am not 100% sure whether the OTel error is causing/caused by/just correlated with the async generator failure...to clarify I also have the Flask app instrumented so there is a non-zero possibility this is a Flask+Anthropic instrumentation interaction issue.

[alexmojaki]

Yes I'm saying it's probably caused entirely by the flask instrumentation. And it doesn't seem to require an actual error to cause the log to appear.