Currently tracing-distributed only sends traces that have an explicit parent span. This is hardcoded, the comments are rather clear and it silently ignores when an out of span event is attempted to be recorded (which was hard to find) .
Yet my opinion is that this is a leftover from when the projects were one. I think the place to decide whether something is worth logging or not is a per-layer filter and should be exposed to the user that way.
Dragging a span around just for the sake of making the layer work is a cumbersome workaround.
Do you think this requirement could be relaxed? Maybe this would require some kind of pseudo parent-id when it's in the root span. Overall this change in behavior would make the whole crate more flexible. If you're not opposed in general, I'll try to help.
Currently tracing-distributed only sends traces that have an explicit parent span. This is hardcoded, the comments are rather clear and it silently ignores when an out of span event is attempted to be recorded (which was hard to find) .
https://github.com/eaze/tracing-honeycomb/blob/5a41b1354bcd76af6a436b5853e6217ba1be2e7f/tracing-distributed/src/telemetry_layer.rs#L214
Yet my opinion is that this is a leftover from when the projects were one. I think the place to decide whether something is worth logging or not is a per-layer filter and should be exposed to the user that way. Dragging a span around just for the sake of making the layer work is a cumbersome workaround.
Do you think this requirement could be relaxed? Maybe this would require some kind of pseudo parent-id when it's in the root span. Overall this change in behavior would make the whole crate more flexible. If you're not opposed in general, I'll try to help.