Traces: APM table

[joey-grafana]

The APM table adds application performance management functionality above the traces view service graph, and is based on discussions/improvements with the Tempo team.

This PR also includes a fix to allow the service graph (and so also the APM table) to be viewed in a dashboard.

Special notes for your reviewer:

1. You can use this Gist to setup your dev env to run the table
2. Enable tempoServiceGraph & tempoApmTable in your custom.ini

[grafanabot]

Drone build failed: https://drone.grafana.net/grafana/grafana-enterprise/18205

[grafanabot]

Drone build failed: https://drone.grafana.net/grafana/grafana-enterprise/18438

[brendamuir]

Activate?

[grafanabot]

Drone build failed: https://drone.grafana.net/grafana/grafana-enterprise/21415

[kvrhdn]

I'm not able to recreate the value shown in the rate column, the shown value seems to be almost double what I would expect (see screenshot). Besides using $__range and @ end(), is the query used by the table different?
Weirdly enough, I don't have this issue with the error rate column. Values align with my direct promql query.

Looking at this again, the values always seem to match now. Not sure if you fixed the query or my data was just weird.

with the above in mind, would you still like me to update the rate col name? Also, should I update the error rate col name too?

I'm not sure if there is any precedent in other areas in Grafana? A rate is usually expressed as x per seconds, so it might not be necessary.

[kvrhdn]

Looks good to me! 🚀

[kvrhdn]

I have some unfortunate news though: we discovered that the metrics emitted by Tempo have a slightly different name from those from the agent and the OTel Collector. Our original intention was to align metric names and labels so it would be easy to use both, but somehow I still managed to use different names in Tempo 🙃

We will be correcting this in Tempo (breaking change): grafana/tempo#1478
The following metrics will change:

• traces_spanmetrics_duration_seconds_bucket -> traces_spanmetrics_latency_bucket
• traces_spanmetrics_duration_seconds_count -> traces_spanmetrics_latency_count
• traces_spanmetrics_duration_seconds_sum -> traces_spanmetrics_latency_sum

And the span_status label will become status_code.

I think the current PR uses the old metric names, we should switch this by the time this feature is released (does not have to be part of this PR).