[processor/transform] Add the capability to scale a metric value

[gabrielgiussi]

Component(s)

processor/transform

Is your feature request related to a problem? Please describe.

Currently the metricstransformer already supports scaling a value and a recent PR even extended this for histograms.

Since processor/transformer will eventually replace metricstransformer could we add this feature?

If you agree with feature request I could work on this.

Describe the solution you'd like

Scale a value down/up, even histograms, e.g. to transform a histogram in milliseconds to seconds

scale(0.001) where metric.name == "envoy_vhost_vcluster_upstream_rq_time"

Describe alternatives you've considered

No response

Additional context

No response

[github-actions]

Pinging code owners:

• processor/transform: @TylerHelmuth @kentquirk @bogdandrutu @evan-bradley

See Adding Labels via Comments if you do not have permissions to add labels yourself.

[TylerHelmuth]

A solution for this may have some overlap with #16067

[TylerHelmuth]

@gabrielgiussi yes the transform processor should be able to handle this. I think we can make a generic function that takes a float or int and scales it based on a factor. Instead of scale setting the value itself, though, I think it would work better as a Factory Function (a function that modifies data and then returns it) would be best so that the result can be used anywhere.

Function signature could look like Scale(getter, factor). The we could do things like

• set(value_double, Scale(value_double, 10))
• set(attributes["my-number"], Scale(value_double, 100))
• set(start_time_unix_nano, Scale(start_time_unix_nano, .0001) <- I'm really interested in this implication. I think it would screw with stuff bc the start_time_unix_nano setter expects a nano time.

[github-actions]

This issue has been inactive for 60 days. It will be closed in 60 days if there is no activity. To ping code owners by adding a component label, see Adding Labels via Comments, or if you are unsure of which component this issue relates to, please ping @open-telemetry/collector-contrib-triagers. If this issue is still relevant, please ping the code owners or leave a comment explaining why it is still relevant. Otherwise, please close it.

Pinging code owners:

• processor/transform: @TylerHelmuth @kentquirk @bogdandrutu @evan-bradley

See Adding Labels via Comments if you do not have permissions to add labels yourself.

[github-actions]

This issue has been closed as inactive because it has been stale for 120 days with no activity.

[TylerHelmuth]

@gabrielgiussi OTTL supports arithmetic operations which should handle this use-case:

set(value_double, value_double*0.001) where metric.name == "envoy_vhost_vcluster_upstream_rq_time"

I do think there is a limitation with scaling histograms.

[bacherfl]

If this ticket is available, I would like to work on this @evan-bradley FYI

[bacherfl]

@TylerHelmuth @evan-bradley just a quick question for clarification - Since arithmetic operations can be used for achieving this, as suggested here, is the goal now to either:

• have the arithmetic operations also support handling Histograms, or to
• implement the Scale factory function that can handle both scalar values and histograms?

[TylerHelmuth]

@bacherfl thanks for looking at this.

I am open to opinions on whether we just need histogram support or a new function.

I believe a new function would provide the most consistent workflow for end users, and it could also optionally allow updating the metric unit.

[evan-bradley]

I would also agree we should implement a function.

I would be in favor or either a scale_metric Editor that only operates on all data points within a metric, or a Scale Converter that works on both whole metrics and on individual data points. The Converter would probably be a little more verbose in general usage (set(data_points, Scale(data_points, 10))), but would allow for things like scaling a misadjusted data point if that's ever an issue.

A Scale converter could be good for adjusting individual integer/float values, but I think relying on arithmetic for that is okay.

[TylerHelmuth]

Operating on all the datapoints of a metric is the most similar to the metricstransformprocessor

[bacherfl]

Thanks for your replies @TylerHelmuth @evan-bradley - I implemented this as a new converter function now - feel free to check out the linked PR :)