Proposal: Use Prometheus labels in stats_over_http plugin output

[itsh3sam]

I'm considering contributing a PR to change the Prometheus output in the stats_over_http plugin to use metric labels instead of flattening all metric names. This would provide a more standard Prometheus format and make metric parsing and aggregation much easier for users and downstream tools.

Is it acceptable to prepare a PR for this change? Are there any reasons this approach wouldn't be welcome (compatibility concerns, past discussions, etc.)? If so, should it be opt-in or the default format?

[masaori335]

Sounds nice, but did you check the /_stats/prometheus output format? It's different from your proposal?

https://docs.trafficserver.apache.org/admin-guide/plugins/stats_over_http.en.html#output-format

[itsh3sam]

Yes, I’ve reviewed the current /_stats/prometheus format and docs.
Instead of encoding dimensions into the metric name (e.g. proxy_process_http_303_responses or proxy_process_api_time_10ms), I’m suggesting using Prometheus labels:

proxy_process_http_responses{status="303"}
proxy_process_api_time{le="10"}

This follows Prometheus best practices, avoids metric name explosion, and makes querying and aggregation much easier.

[masaori335]

Cool. Then, the problem is compatibility. ATS employs Semantic Versioning, so this new Prometheus format must be an option for ATS v10.x. We can make it default by ATS v11.0.0 if we want.