Ping @evan-bradley @TylerHelmuth @bogdandrutu @edmocosta

This seems to break the Histogram data model, particularly removing sum.

Why not (when removing buckets) place the removed data into some single bucket or 'overflow' bucket?

I'm not really sure I understand the use case at all, but I'd be afraid this would break many metric backends and generally be a large footgun. Can you describe that more either in this PR or in the issue?

Why not (when removing buckets) place the removed data into some single bucket or 'overflow' bucket?

I'm not sure about that solution, to be honest. One of the benefits is to reduce unused data.

I'm not really sure I understand the use case at all, but I'd be afraid this would break many metric backends and generally be a large footgun.

Just if they use it and don't know what they are doing (as the rest of features we have in the OpenTelemetry Collector). Don't see the issue adding more flexibility.

Can you describe that more either in this PR or in the issue?

It is common for Prometheus users doing this:

metric_relabel_configs:
  - source_labels: [__name__, le]
    regex: 'example_latency_seconds_bucket;0\.0.*'
    action: drop

I want to replicate this feature in OpenTelemetry Collector.

One ref, for instance https://www.robustperception.io/why-are-prometheus-histograms-cumulative/

Ah, so if you're removing buckets prometheus style, you have the benefit that buckets are just less than a threshod, I.e. they include counts from previous buckets.

If you do the same in opentelemetry you NEED to add the counts from previous buckets into future buckets or you have literally lost data. I.e. if you use this function as it is defined, you just have a broken histogram, you wouldn't wind up with the same value as prometheus dropped buckets gives you.

At a minimum you should:

• Shift bucket counts to nearest bucket (either up or down)
• Expand bucket boundaries or nearest bucket to incorporate the lost bucket.

Then you'd get the prometheus behavior.

• Shift bucket counts to nearest bucket (either up or down)

Any idea about how to do this? Because per my understanding
#40281 (comment)

Thanks for your comments @jsuereth

This PR was marked stale due to lack of activity. It will be closed in 14 days.

Closed as inactive. Feel free to reopen if this PR is still being worked on.