A Span which represents a time-period and encloses events. Spans are hierarchical -- i.e., they form a tree.
A Trace which is essentially a "root" Span with some additional metadata.
This should be roughly compatible with OpenTelemetry even though the goal is to keep it extremely minimal as far as possible. The most interesting part will be evolving the Event type to be a discriminated union of very opinionated event types which correspond to things happening in the ML app lifecycle.
I added a simple console provider for this API, tied standard Python logger to this API as well (for unstructured logging). We may or may not keep unstructured logging around but it is a start.
A simple Telemetry API with three basic notions:
This should be roughly compatible with OpenTelemetry even though the goal is to keep it extremely minimal as far as possible. The most interesting part will be evolving the Event type to be a discriminated union of very opinionated event types which correspond to things happening in the ML app lifecycle.
I added a simple
consoleprovider for this API, tied standard Python logger to this API as well (for unstructured logging). We may or may not keep unstructured logging around but it is a start.Sample output: