@csmarchbanks
Thanks for your reply!
My use case is using Heatmap in Grafana via a Histogram in Prometheus, I can only get the sum of each bucket when I use Histogram, because the data source is from another module which I can't invasive modify too much, so I have to update the metrics like:

age_to_count = ... # another module
for age, count in age_to_count:
  for _ in range(count):
    age_to_count_hist.observe(age)

It's a bit of waste because every observe is O(n) to find the corresponding bucket.

@csmarchbanks Or we have to write code like this which is not safe to access the protected fields:

age_to_count = ... # another module
for age, count in age_to_count:
  age_to_count_hist._sum.inc(age * count)
  for i, bound in enumerate(age_to_count_hist._upper_bounds):
    if age <= bound:
      age_to_count_hist._buckets[i].inc(count)
      break

My thought for this case is to use a custom collector, specifically creating the buckets (as you somewhat already have those). See

That will allow at scrape time for you to call age_to_count and generate the histogram directly. I think with your current implementation you will also be "re-observing" many of your buckets if you are calling inc(age*count) on some sort of timer.

Let me know what you think of that suggestion!