Full Prometheus metric name normalization disabled by default

[swiatekm]

Prometheus and otel metric naming conventions aren't entirely compatible. Prometheus metric name normalization is a feature intended to convert metrics between the two conventions. See the upstream documentation for details.

In the 0.76.1 release, the feature flag for this normalization was enabled by default.

The feature primarily affects the prometheus and prometheusremotewrite exporters. However, it also modifies some metrics collected by the prometheus receiver. More specifically, it trims certain suffixes from metric names. Unfortunately, this affects a lot of widely used metrics. For example, the standard container CPU usage metric:

container_cpu_usage_seconds_total -> container_cpu_usage_seconds

This change breaks a lot of content built using existing metric names and prevents the Prometheus receiver from being used as a drop-in replacement for Prometheus. Therefore, we've decided to default to having this flag disabled. #1138 implements this decision.

The behavior can be re-enabled by passing the --feature-gates=+pkg.translator.prometheus.NormalizeName flag to the collector at startup.

There is an ongoing discussion about making this behavior configurable at runtime. Please follow this issue if you'd like to learn more.

[swiatekm]

Unpinning this, as the change was reverted upstream in open-telemetry/opentelemetry-collector-contrib#23229. I'm going to keep it open for now.

[swiatekm]

Closing, as this is not relevant anymore.