Tracing decorator for functions

[hooverdc]

Is your feature request related to a problem?
Tracing every use of a function is prone to error. It is very hard to do in some contexts, such as functions from a third party library that don't integrate with OTel, without lots of manual span management.

Describe the solution you'd like
I would like to have a first party tracing decorator.

Describe alternatives you've considered
I have rolled my own for now:

import functools
from contextlib import nullcontext
from typing import Callable

from opentelemetry import trace
from opentelemetry.trace import NonRecordingSpan


def traced(name: str) -> Callable:
    """Wrap a function in an OTel span."""

    def decorator(func: Callable):
        @functools.wraps(func)
        def wraps(*args, **kwargs):
            if isinstance(trace.get_current_span(), NonRecordingSpan):
                span = nullcontext()
            else:
                span = trace.get_tracer(func.__module__).start_as_current_span(name)
            
            with span:
                return func(*args, **kwargs)

        return wraps

    return decorator

Additional context
Honeycomb does something similar in their beeline module with the traced decorator.

[srikanthccv]

You could achieve same with this @tracer.start_as_current_span(name).

@tracer.start_as_current_span("name")
def say_hello():
  print("Hello!")

if __name__ == "__main__":
  hello()

[aabmass]

We should probably clarify this in our documentation, at least in the API doc

opentelemetry-python/opentelemetry-api/src/opentelemetry/trace/__init__.py

Line 343 in faf8868

"""Context manager for creating a new span and set it

[Cruuncher]

@aabmass What is @tracer here? Is tracer the return from trace.get_tracer(...).

A convenience might be better around this, as it seems like the one-line decorator here would be
@trace.get_tracer(__name__).start_as_current_span("some_name") which is very verbose for something that seems like it should be a common use case

[aabmass]

Is tracer the return from trace.get_tracer(...).

Yes. Typically you would hold the tracer as a package, module, or class level variable and use that throughout your code. We don't want to make any assumptions about what the tracer's name is.

[s4v4g3]

I had the same need so I created this extension package for OpenTelemetry:
https://pypi.org/project/otel-extensions/

Very much open to feedback but would love to equivalent features integrated into the opentelemetry-api or opentelemetry-sdk library.