Standardize prometheus naming convention for keda metrics

[pragmaticivan]

Proposal

Because of issues with Prometheus remote writers, it's ideal that we follow Prometheus suffix conventions for counters. If not followed strictly, external vendors supporting Prometheus remote writers might receive the metric as a Gauge instead.

Context:

• https://prometheus.io/docs/practices/naming/
• https://docs.newrelic.com/docs/infrastructure/prometheus-integrations/install-configure-remote-write/set-your-prometheus-remote-write-integration/#mapping
• Consider (optionally?) enforcing certain metric naming conventions prometheus/client_golang#725
• https://github.com/thanos-io/thanos/pull/3992/filesPrometheus

Use-Case

Vendors such as newrelic can use Prometheus metrics via remote writer properly.

Is this a feature you are interested in implementing yourself?

Maybe

Anything else?

No response

[pragmaticivan]

We might need to deprecate existing metrics and create new ones.

[pragmaticivan]

@JorTurFer, if approved, I can probably take this one.

[JorTurFer]

WDYT @kedacore/keda-core-contributors ?

[zroubalik]

@pragmaticivan could you please write down the list of metrics to be changed. The old and new name. Thanks!

[pragmaticivan]

@zroubalik afaik, only the ones marked as red.

Although I was thrown off by naming some being gauge instead of a histogram.

[pragmaticivan]

I also noticed some of the HELP (descriptions) don't match the description in the doc.

[pragmaticivan]

Possible steps:

1. Update Description for affected metrics to include a prefix like: Deprecation Warning - This metric is deprecated and replaced with 'new_metric'
2. Create new metrics and add increments right below the code for the existing ones.
3. Update docs (include deprecation notice)
4. Update Grafana template

[tomkerkhove]

Sounds like good execution steps imo, are you willing to contribute this?

For 2), let's make sure we increment new metrics first though.

[pragmaticivan]

Yes, I can take that one.

[pragmaticivan]

After taking a deeper look yesterday, keda_scaler_errors vs keda_scaler_errors_total seems to be the same metric, called in the same place. The only difference is that keda_scaler_errors has the labels in place, but technically you could use this one with the same intent (sum) so I don't think we need the counter for all since we already have the other metric.

[pragmaticivan]

This issue might be obsolete if this is available: https://github.com/kedacore/keda/pull/4860/files 🎉

[tomkerkhove]

Well not really because Prometheus metrics are not being removed; we're adding it next to each other :)

[wonko]

I made a PR with a (biased) suggestion to rework the metric names.

• units are added (to latency)
• some wordings have been changed in the help
• some metrics have a new name
• deprecated a sum() metric
• every metric that needs a deprecation has been duplicated to provide the new metric

Open to suggestions.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.

[stale]

This issue has been automatically closed due to inactivity.

[JorTurFer]

This is in progress