fix(core): use compensated summation for histograms#1666
Merged
gh-worker-dd-mergequeue-cf854d[bot] merged 1 commit intoMay 18, 2026
Merged
Conversation
Use modified Neumaier algorithm to calculate sums, counts, and quantiles
for histogram samples.
The naive `sum += value * weight` loop suffers catastrophic cancellation
when the sample stream contains values of wildly different magnitudes.
The classic Kahan/Peters counter-example `{1, +1e100, 1, -1e100}`
evaluates to 0 with naive summation but to the correct 2.0 with the new
algorithm.
Signed-off-by: Mark Kirichenko <mark.kirichenko@datadoghq.com>
dd-octo-sts
Bot
added
area/core
tobz
changed the title
tobz
approved these changes
May 15, 2026
Binary Size Analysis (Agent Data Plane)Target: 55bca14 (baseline) vs df02eb3 (comparison) diff
|
| Module | File Size | Symbols |
|---|---|---|
core |
-33.99 KiB | 1451 |
smallvec |
+22.57 KiB | 62 |
figment |
-10.78 KiB | 12 |
saluki_core::data_model::event |
-8.63 KiB | 22 |
[Unmapped] |
-4.36 KiB | 1 |
anon.0cac25fc52ba6f4fc475348a8c66d8e3.39.llvm.4081512695721216103 |
+3.90 KiB | 1 |
anon.8bb1cfcdf181d421e8889bc3626b8144.17.llvm.125851317713716366 |
-3.81 KiB | 1 |
[sections] |
-3.30 KiB | 7 |
anon.bcfbe2edddf7aafb2d5d5e0cc5ffa1e5.16.llvm.12024661772337668300 |
-3.15 KiB | 1 |
hashbrown |
+3.15 KiB | 24 |
anon.0f2fa1d1fad1031510176699744ee20b.644.llvm.3708903403341574001 |
+3.06 KiB | 1 |
papaya |
+2.56 KiB | 11 |
anon.eb51c975e2567ebfca80d7da0abd4cd1.7.llvm.11556325092004374934 |
-2.55 KiB | 1 |
anon.40340ab29c26454e228b882d4ecb70d4.0.llvm.14605026808982526804 |
+2.54 KiB | 1 |
saluki_api::DynamicRoute::http |
+1.99 KiB | 1 |
anon.bdadb00871588c874a55b8b73ce579a9.0.llvm.8997752761958168434 |
-1.93 KiB | 1 |
anon.41b0b8befc3118c3a2b0f17ec06872eb.9.llvm.17862187855904175761 |
+1.93 KiB | 1 |
alloc |
-1.91 KiB | 68 |
serde_core |
+1.88 KiB | 40 |
tokio_util |
-1.84 KiB | 11 |
Detailed Symbol Changes
FILE SIZE VM SIZE
-------------- --------------
[NEW] +6.41Ki [NEW] +6.33Ki matchit::router::Router<T>::insert::h261aa988f7ce6a25
[NEW] +6.14Ki [NEW] +6.05Ki matchit::router::Router<T>::insert::hfe75556626a71a8d
[NEW] +6.07Ki [NEW] +5.99Ki matchit::tree::Node<T>::insert::h889792c8a42776b1
+904% +5.75Ki +10e2% +5.75Ki axum::routing::Router<S>::route::h0531592b945792f9
[NEW] +5.42Ki [NEW] +5.33Ki serde_core::de::MapAccess::next_value::ha8691cfa6c85eadd
+795% +5.40Ki +908% +5.40Ki tokio::runtime::runtime::Runtime::block_on::h59580466f3413de1
+992% +5.27Ki +12e2% +5.27Ki tokio::runtime::runtime::Runtime::block_on::h26cc5fddc7a2e095
[NEW] +3.90Ki [NEW] +16 anon.0cac25fc52ba6f4fc475348a8c66d8e3.39.llvm.4081512695721216103
[NEW] +3.06Ki [NEW] +74 anon.0f2fa1d1fad1031510176699744ee20b.644.llvm.3708903403341574001
[DEL] -2.55Ki [DEL] -80 anon.eb51c975e2567ebfca80d7da0abd4cd1.7.llvm.11556325092004374934
[DEL] -2.89Ki [DEL] -2.77Ki quick_cache::shard::CacheShard<Key,Val,We,B,L,Plh>::insert::haf55a3f4a81ec55a
[DEL] -3.15Ki [DEL] -74 anon.bcfbe2edddf7aafb2d5d5e0cc5ffa1e5.16.llvm.12024661772337668300
[DEL] -3.48Ki [DEL] -2.33Ki _<serde_core::de::impls::<impl serde_core::de::Deserialize for core::time::Duration>::deserialize::DurationVisitor as serde_core::de::Visitor>::visit_map::hed68c5e9e330be6d
[DEL] -3.81Ki [DEL] -16 anon.8bb1cfcdf181d421e8889bc3626b8144.17.llvm.125851317713716366
-51.3% -4.36Ki [ = ] 0 [Unmapped]
[DEL] -5.26Ki [DEL] -5.10Ki _<figment::value::de::ConfiguredValueDe<I> as serde_core::de::Deserializer>::deserialize_struct::hedaa6bc6b4faff3a
[DEL] -6.01Ki [DEL] -5.93Ki matchit::tree::Node<T>::insert::ha63513d725e691c6
[DEL] -6.11Ki [DEL] -6.02Ki matchit::router::Router<T>::insert::h02e8e157485e9a5b
[DEL] -6.17Ki [DEL] -6.06Ki axum::routing::path_router::PathRouter<S,_>::route::h75fd76341dec4a4b
[DEL] -6.39Ki [DEL] -6.31Ki matchit::tree::Node<T>::insert::h2e5068899f601750
-0.6% -37.1Ki -0.6% -25.2Ki [7148 Others]
-0.1% -39.9Ki -0.1% -19.6Ki TOTAL
Regression Detector (Agent Data Plane)Run ID: Optimization Goals: ✅ No significant changes detectedFine details of change detection per experiment (35)Experiments configured
Bounds Checks: ✅ Passed (5)
ExplanationA change is flagged as a regression when |Δ mean %| > 5.00% in the regressing direction for its optimization goal AND SMP marks the experiment as a regression ( |
atanzu
changed the title
gh-worker-dd-devflow-36fce6
Bot
added
mergequeue-status: queued
mergequeue-status: in_progress
and removed
mergequeue-status: queued
labels
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: df02eb32c4
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| let mut ddsketch = DDSketch::default(); | ||
| for sample in histogram.samples() { | ||
| ddsketch.insert_n(sample.value.into_inner(), sample.weight); | ||
| ddsketch.insert_n(sample.value.into_inner(), sample.weight.0 as u64); |
There was a problem hiding this comment.
When this encoder handles a histogram built from a non-integer sample rate such as @0.21, Histogram::insert now stores sample.weight as the raw weight (~4.76), while summary.count() rounds that to the nearest sample count. This cast truncates the same sample to 4 before inserting it into the DDSketch, so encoded histogram payloads undercount fractional-weight samples and can disagree with the aggregate count/sum produced from the same histogram. Convert the raw weight using the same rounding/accounting policy before passing it to insert_n.
Useful? React with 👍 / 👎.
gh-worker-dd-mergequeue-cf854d
Bot
merged commit 19791e3
into
main
80 of 82 checks passed
gh-worker-dd-devflow-36fce6
Bot
added
mergequeue-status: done
and removed
mergequeue-status: in_progress
labels
dd-octo-sts Bot
pushed a commit
that referenced
this pull request
May 18, 2026
## Summary
Use modified Neumaier algorithm to calculate sums, counts, and quantiles for histogram samples.
The naive `sum += value * weight` loop suffers catastrophic cancellation when the sample stream contains values of wildly different magnitudes. The classic Kahan/Peters counter-example `{1, +1e100, 1, -1e100}` evaluates to 0 with naive summation but to the correct 2.0 with the new algorithm.
## Change Type
- [x] Bug fix
- [ ] New feature
- [ ] Non-functional (chore, refactoring, docs)
- [ ] Performance
## How did you test this PR?
Added unit tests to check correctness.
## References
[Similar PR in Datadog-Agent](DataDog/datadog-agent#49913).
Co-authored-by: mark.kirichenko <mark.kirichenko@datadoghq.com> 19791e3
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Use modified Neumaier algorithm to calculate sums, counts, and quantiles for histogram samples.
The naive
sum += value * weightloop suffers catastrophic cancellation when the sample stream contains values of wildly different magnitudes. The classic Kahan/Peters counter-example{1, +1e100, 1, -1e100}evaluates to 0 with naive summation but to the correct 2.0 with the new algorithm.Change Type
How did you test this PR?
Added unit tests to check correctness.
References
Similar PR in Datadog-Agent.