Log errors

[dmathieu]

Area(s)

area:error

Is your change request related to a problem? Please describe.

Logs have semantic conventions for exceptions, but not all errors are exceptions (some languages such as Go don't even have the notion of exception).

Describe the solution you'd like

Introduce the error namespace for logs, and add the field error.message to that namespace.

Describe alternatives you've considered

We could also not introduce a new namespace, and document that non-exception errors should be treated as exceptions.

Additional context

• Log Exceptions
• Error namespace

[pyohannes]

There's been some related discussions in open-telemetry/opentelemetry-specification#3198.

I think it was the discussion in this issue that resulted in the introduction of the error namespace.

Also, in this issue there were some interesting discussions about the fact that the term "error" corresponds to the span status in tracing, but to the severity in logging. For spans, it seemed feasible to use the status description for capturing error messages, however, it wasn't clear how this should translate to other signals.

[joaopgrassi]

I wonder if discussions about this should be kept in the spec issue or here. Seems it goes beyond semantic conventions and affects other areas as well.

[dmathieu]

With exception being stabilized, I suppose a rename of exception into error is not possible.
The specification above did mention having both exceptions and errors though.

I'm also happy opening a spec issue if that'd be a better place.

[pellared]

I would say that for Go we need error to differ between regular error handling and panics where we could use exception.

[pellared]

@dmathieu, with https://github.com/open-telemetry/opentelemetry-go/blob/e3892f062f3f92cce9b2a61478c46babe6f6331d/sdk/trace/span.go#L448-L460 I think we are fine using exception.type and exception.message and follow https://github.com/open-telemetry/semantic-conventions/blob/main/docs/exceptions/exceptions-logs.md in Log Bridges. WDYT?

[dmathieu]

Yes, that's the alternative suggested in the issue description. I'm fine going with that alternative.

[dmathieu]

Closing per the last 2 comments.