trace() fails to execute block when zipkin disabled or not initialized.

[ParensAllTheWayDown]

I'm trying to ensure that our fastapi application continues to execute instrumented block if zipkin is/isnot configured. In the 'isnot' scenario, zipkin trace() scope function fails to execute the scope because get_tracer() returns a None and subsequent call to tracer.new_child fails.

    def __enter__(self) -> "trace":
        tracer = get_tracer()
        parent = _cur_span_ctx_var.get()
        if parent is None:
            parent = get_root_span()
        self._span = tracer.new_child(parent.context)
        self._tok = _cur_span_ctx_var.set(self._span)
        self._span.__enter__()
        self._span.name(self._name)
        self._span.kind(self._kind)
        return self

Would it be reasonable to fall back to an empty __enter__ & __exit__ if there is no tracer available? That would allow the instrumented code to run. Perhaps a warning message that indicates that things are not proceeding as expected.

I can work around this, but I have to duplicate the instrumented scope in a try-except in order to ensure that the instrumented code is executed in the case that zipkin configuration has been intentionally skipped. For example:

    @property
    def userid(self):
        try:
            with trace('ad_provider', kind='SERVER'):
                return self.get_user_from_validation_info()
        except AttributeError as ae:
            return self.get_user_from_validation_info()

[mchlvl]

Hey @ParensAllTheWayDown ! Thanks for the suggestion! It sounds reasonable to me, so I prepared #36 that might help with this.

Just to make sure about the use cases - am I correctly assuming this happens when you do not add the middleware (that is skip app.add_middleware(ZipkinMiddleware)), or is there another situation when this happens?

[ParensAllTheWayDown]

Short answer 'Yes'.

Longer answer follows: We are using 'Settings' subclass to define all settings for the application. If the zipkin subsection is not included, then ZipkinMiddleware is not included in our middleware stack.

class ZipkinSettings(BaseModel):
    """
    Configure Zipkin middleware layer
    """
    host: str = Field('127.0.0.1', description="The zipkin database that will collect call metrics.")
    port: int = Field(9411, description='The port to send performance metrics.')
    service_name: str = Field('gdmlapi', description='the name that identifies this gdmlapi server.')
    sample_rate: float = Field(1.0)
    inject_response_headers: bool = Field(True)
    force_new_trace: bool = Field(False)
    json_encoder: Callable = Field(json.dumps)
    header_formatter = Field(B3Headers)

and Settings subclass:

class Settings(BaseSettings):
    ...
    zipkin: Optional[ZipkinSettings]
    ...
    class Config:
        env_nested_delimiter = '__'

Our middleware include/exclude logic includes this:

        if settings.zipkin:
            config = ZipkinConfig(
                host=settings.zipkin.host,
                port=settings.zipkin.port,
                service_name=settings.zipkin.service_name,
                sample_rate=settings.zipkin.sample_rate,
                inject_response_headers=settings.zipkin.inject_response_headers,
                force_new_trace=settings.zipkin.force_new_trace,
                json_encoder=settings.zipkin.json_encoder,
                header_formatter=settings.zipkin.header_formatter
            )
            logger.debug(f"Adding zipkin: {config}")
            app.add_middleware(ZipkinMiddleware, config=config)

[mchlvl]

Alright, thanks for clarifying, will make a new version with this later today 🙌

[mchlvl]

@ParensAllTheWayDown Just merged and released 0.2.4. Let me now if this helps your usecase!

[ParensAllTheWayDown]

Thank you for taking this into consideration. I'll give it a try tomorrow.