improve tracing for openai agents #205
Conversation
|
Note that we appear to have a tracing bug where some spans don't get closed. We will address this as a separate issue. |
| trace_id = gen_trace_id() | ||
| with trace("Orchestrator evaluator", trace_id=trace_id): |
There was a problem hiding this comment.
Does trace ID have to be explicitly generated here? It is the default correct? (this calls our deterministic UUID generator, and have it on our TODO to stop using the RNG shared with the user so we don't affect their RNG use with our own)
| instructions="You only respond in haikus.", | ||
| ) | ||
| trace_id = gen_trace_id() | ||
| with trace("Hello World", trace_id=trace_id): |
There was a problem hiding this comment.
I was under the impression samples like these were intentionally meant to look like https://github.com/openai/openai-agents-python/blob/main/examples/basic/hello_world.py for comparison. OpenAI chooses not to demonstrate tracing in their hello world, is it important that Temporal demonstrate OpenAI tracing in the same sample? Should we upstream a PR to OpenAI samples if we field demonstrating tracing in hello world is important? I think it may be clearer if we are wanting to demonstrate something beyond just showing OpenAI sample equivalents, we do it in a separate sample.
What was changed
Added tracing to all of the OpenAI Agents examples.
Why?
Some examples had incomplete tracing.
Checklist
Closes
How was this tested:
Manual validation via OpenAI platform UI.
Any docs updates needed?
N/A