Null Input & Output when calling Azure OpenAI model

[haroldpijpelink]

Hi there everyone,

I have this issue where I can run promptflow and get the outputs I want, but tracing does not seem to work properly. Instead of getting a success or fail status, it says the following in the output json:
status: {
status_code: "Unset"
description: ""
}
Moreover, the Input and Output fields show as Null in the UI and I have no information on the amounts of tokens used.

I get this both when I use start_trace() in the tracing docs and when I try to access through the Prompty extension. I realize that this might be some issue with OpenTelemetry settings and not Promptflow itself, but if anyone could point me in the right direction that'd be great :)

[Flowhill]

Have you set

environment_variables:
  PF_DISABLE_TRACING: "false"

In your flow.dag.yaml?

I'm getting the same issue when I'm running it through the python file itself (via fastapi). Then there is no neat trace.
I do get proper traces when I click the run button in the flow.dag.yaml with the promptflow extension.

[Flowhill]

@haroldpijpelink
I was able to solve this for myself.
By reverting to promptflow-tracing==1.17.0 (and thus promptflow itself as well) I did get the proper tracing on flows started with flow-as-a-function.

e.g.

f = load_flow(<path/to/flow/folder>)
        result = f(
            <input-mapping>
        )

I don't use start_trace() as I use the flow-as-a-function.

This seems to break as soon as you update to promptflow(-tracing)>=1.17.1