Influx metrics: Errors aren't always errors

[franzliedke]

The Influx metric currently assumes that the presence of errors attached to a trace means that the trace itself failed.

In our production environment, we are also attaching temporary failures, e.g. when we retry service requests, to the trace. I guess we need new APIs to distinguish trace error state and other "failures".