Support Prometheus native histograms

[runiq]

A note for the community

• Please vote on this issue by adding a 👍 reaction to the original issue to help the community and maintainers prioritize this request
• If you are interested in working on this issue or have submitted a pull request, please leave a comment

Use Cases

Compared to Prometheus' current histogram implementation, native histograms:

• use a uniform bucket sizing scale
• remove the requirement to know your bucket sizes beforehand
• allow performing arithmetic on different bucket sizings (by using the less granular size as the "baseline")
• save only one time series per histogram (not one per bucket), which can save a lot of space

Attempted Solutions

No response

Proposal

No response

References

https://www.youtube.com/watch?v=AcmABV6NCYk&list=PLoz-W_CUquUmVOjYTqWHX4CJ0jP61vif8&index=2

Version

vector 0.27.0 (x86_64-unknown-linux-gnu 5623d1e 2023-01-18)

[jszwedko]

Note that we'll probably want to wait for this feature to stabilize in Prometheus before adding to Vector. It currently appears to be undocumented.

[jacobstr]

To follow up @jszwedko's comment, indeed this un/under-documentation still appears to be the case in April 2024. It looks like the functionality is making it's way into mainline prom in fits and starts but:

• The only blurb in the official docs is that it's experimental as of 2.4.0
• Most of the other "documentation" are conference talks e.g:
  • Ganesh Vernekar Grafana Labs, June 1 2022
  • Bjorn Rabenstein Grafana Labs, April 20, 2023
• Project milestones on Github show 72% completion

Much (all?) of this development appears to mostly center around the golang prometheus community.