Prometheus exporter: 0.154.0 adds metric suffixes despite `add_metric_suffixes: false`

[svifred]

Component(s)

No response

What happened?

Describe the bug
Using Prometheus exporter with add_metric_suffixes: false in 0.154.0 adds unit to metric name, where 0.153.0 using the same config does not.

I suspect, but have not verified, that prometheus/common#877 pulled in by open-telemetry/opentelemetry-collector#15379 is buggy.

Steps to reproduce
Define config for opentelemetry-collector as below and export a metric, say

http.server.requests{method=GET,outcome=SUCCESS,status=200,uri=/yourUrl} count=1 sum=1108.618208 le10=0 ... le5000=0 0

Then curl the /metrics endpoint.

What did you expect to see?
Using 0.153.0 I expect and do get:

# HELP http_server_requests HTTP server request processing time
# TYPE http_server_requests histogram
http_server_requests_bucket{job="unknown_service",method="GET",otel_scope_name="",otel_scope_schema_url="",otel_scope_version="",outcome="SUCCESS",status="200",uri="/yourUrl",le="10"} 0

What did you see instead?
Using 0.154.0 I expect the metric name to remain http_server_requests_bucket, but I get:

# HELP http_server_requests_milliseconds HTTP server request processing time
# TYPE http_server_requests_milliseconds histogram
http_server_requests_milliseconds_bucket{job="unknown_service",method="GET",otel_scope_name="",otel_scope_schema_url="",otel_scope_version="",outcome="SUCCESS",status="200",uri="/yourUrl",le="10"} 0

The metric name has had an _milliseconds added.

Collector version

v0.154.0

Environment information

Environment

Docker container running platform linux/amd64 for both versions for debugging locally.

Same images running i Kubernetes v1.29.14+41c4e9b exhibits same issue.

OpenTelemetry Collector configuration

receivers:
  otlp:
    protocols:
      grpc:
        endpoint: 0.0.0.0:4317
      http:
        endpoint: 0.0.0.0:4318

processors:
  batch:

exporters:
  debug:
    verbosity: normal
  prometheus:
    add_metric_suffixes: false
    endpoint: 0.0.0.0:8889

service:
  pipelines:
    traces:
      receivers: [otlp]
      processors: [batch]
      exporters: [debug]
    metrics:
      receivers: [otlp]
      processors: [batch]
      exporters: [debug, prometheus]

Log output

Additional context

No response

Tip

_{React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.}

[prabhaks]

@svifred

This could be the potential root cause - happy to submit a fix for this.

After tracing the regression, the issue originates from the bump of github.com/prometheus/common pulled in via open-telemetry/opentelemetry-collector#15379. The otlptranslator.MetricNamer struct (used in opentelemetry-collector-contrib's exporter/prometheusexporter/collector.go) passes WithMetricSuffixes: false when add_metric_suffixes: false is configured. However, the updated prometheus/common library now applies unit suffixes (e.g. _milliseconds) in a separate path that is not gated by that flag — so the setting is silently ignored for unit suffixes while still suppressing type suffixes.

Workaround (while a fix is in progress): replace add_metric_suffixes: false with the new explicit setting:

exporters:
  prometheus:
    translation_strategy: UnderscoreEscapingWithoutSuffixes
    endpoint: 0.0.0.0:8889

Note: add_metric_suffixes is also deprecated in favor of translation_strategy, so this is the recommended long-term config anyway.

The fix needs to go in opentelemetry-collector-contrib rather than this repo. I can open a PR there to either correct the MetricNamer behavior for the WithMetricSuffixes: false path, or add a regression test to pin the expected behavior.

What do you think?

[prabhaks]

Could someone assign this issue to me? I'd like to work on a fix.

[svifred]

Workaround (while a fix is in progress): replace add_metric_suffixes: false with the new explicit setting:

exporters:
  prometheus:
    translation_strategy: UnderscoreEscapingWithoutSuffixes
    endpoint: 0.0.0.0:8889

Note: add_metric_suffixes is also deprecated in favor of translation_strategy, so this is the recommended long-term config anyway.

[...]

What do you think?

Works beautifully, thank you :)

The option translation_strategy is documented in OpenTelemetry SDK docs in Prometheus Exporter and upstream in Prometheus OpenTelemetry UTF-8 notes, and UnderscoreEscapingWithoutSuffixes is the right option in our case.

I did however expect having to bump opentelemetry-collector to v0.155.0, as 5459acf is listed for v0.155.0. But regardless, changing the config to use translation_strategy works in v0.154.0+.

[github-actions]

Pinging code owners for exporter/prometheus: @Aneurysm9 @dashpole @ArthurSens. See Adding Labels via Comments if you do not have permissions to add labels yourself. For example, comment '/label priority:p2 -needs-triage' to set the priority and remove the needs-triage label.