Using Phoenix for LLM Observability: Debugging Session Issues

Hi, We want to use Phoenix for LLM observability to replace Weave from Weights and Biases. I have a local setup of it running with docker compose just to test it out first. My setup includes the Opentelemetry Collector and then Phoenix (with its Postgres DB). I'm using the Opentelemetry Collector, as that is what we will use in our K8s Cluster to get the data the the Phoenix deployment. I then have a local python script that uses a Pydantic AI agent that does some tool calls. These traces are collected using the Openinference and Opentelemetry and sent to the Opentelemetry Collector and that in turn sends the traces to Phoenix. All of this works and that is great. My problem comes in when trying to use sessions. Our use case is a chatbot and for that sessions will be very useful to our devs when debugging problems. I add a session id to the root trace span using with using_attributes(session_id="should_be_conversation_id", user_id="should_be_user_id"):, and it does create the session. When I open the session in Phoenix where I expect the messages to be it just says undifined. The traces have an input value and an output value (I think this is what the sessions use for this display), so I'm not really sure what I'm doing wrong. Any help on this would be greatly appreciated. (I'll add some screenshots in the thread🧵)