Prometheus: add native histogram support

[howardjohn]

Title: Prometheus: add native histogram support

Description: https://prometheus.io/docs/concepts/metric_types/#histogram

Prometheus has a new native histogram mode which (claims to) make a much better tradeoff between metrics cost and usefulness by having dynamic bucket sizing. It would be great to have support for this in Envoy

[KBaichoo]

cc @jmarantz @ggreenway

[ggreenway]

This sounds interesting. But I couldn't find any documentation on exactly what the format it is, or how it is represented in the exposition format. Can you point me to docs on the format?

[howardjohn]

@ggreenway I was surprised by the lack of clear specification. https://github.com/prometheus/client_golang/pull/1150/files is an example of the Go implementation, but it doesn't link to any specification either and has a lot of seemingly magic stuff.

My understanding is this also depends on protobuf encoded Prometheus encoding, which might be a large pre-req before even getting to the histogram support itself.

Its quite plausible its a bit too early to adopt this in Envoy (or at least not at a 'stable' level)

[ggreenway]

I had thought that prometheus stopped using the protobuf exposition format, which used to be true but is now not true. Found a bit of the backstory here: https://github.com/prometheus/client_model/blob/master/README.md

[euroelessar]

OpenTelemetry has a good documentation for exponential histograms: https://opentelemetry.io/docs/specs/otel/metrics/data-model/#exponentialhistogram As well as a comparison with Prometheus' native histograms: https://opentelemetry.io/docs/specs/otel/compatibility/prometheus_and_openmetrics/#exponential-histograms (spoiler: they are compatible and can be converted to each other)

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had activity in the last 30 days. It will be closed in the next 7 days unless it is tagged "help wanted" or "no stalebot" or other activity occurs. Thank you for your contributions.

[github-actions[bot]]

This issue has been automatically closed because it has not had activity in the last 37 days. If this issue is still valid, please ping a maintainer and ask them to label it as "help wanted" or "no stalebot". Thank you for your contributions.

[howardjohn]

/reopen /help-wanted

[ggreenway]

@howardjohn have there been any updates on the stability of the format, docs, etc?

[beorn7]

I'm currently writing the spec (which hopefully will get marked as final in September). Currently, native histograms can only be exposed via protobuf, so the protobuf spec will already help a lot.

• proto2: https://github.com/prometheus/client_model/blob/master/io/prometheus/client/metrics.proto
• proto3: https://github.com/prometheus/prometheus/blob/main/prompb/io/prometheus/client/metrics.proto

If you are wondering what things mean, you could look at the code of prometheus/client_golang or prometheus/client_java or just ask me. :o)

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had activity in the last 30 days. It will be closed in the next 7 days unless it is tagged "help wanted" or "no stalebot" or other activity occurs. Thank you for your contributions.

[github-actions[bot]]

This issue has been automatically closed because it has not had activity in the last 37 days. If this issue is still valid, please ping a maintainer and ask them to label it as "help wanted" or "no stalebot". Thank you for your contributions.

[adams-shaun]

+1 libcircllhist is great at what it does, but seems to run with high overhead over extended periods of time and large sample counts / numbers of histograms to accumulate.