Rethink client metrics for sampled/unsampled spans/traces

[yurishkuro]

Requirement - what kind of business use case are you trying to solve?

As we're gradually introducing delayed sampling into clients (#1861), our current started_spans/started_traces metrics become less meaningful and do not reflect what happens.

Problem - what in Jaeger blocks you from solving the requirement?

Metrics are currently annotated with a sampled label (yes/no), but when sampling is delayed we don't know if the span/trace will be sampled later. In reverse, finished_spans metric does not have a label for sampled state.

Proposal - what do you suggest to solve the problem or improve the existing situation?

• Introduce a 3rd value delayed for the sampled label on the xxx_started metrics.
• Add sampled label to the xxx_finished metrics (currently it does not have this label).