rust

goodhistogram (Rust)

Rust implementation of github.com/cockroachdb/goodhistogram: a lock-free, Prometheus-aligned exponential histogram.

This crate lives in the same repository as the canonical Go implementation. The two are kept in lockstep by shared conformance fixtures generated from the Go code (../datatest/); tests/datatest.rs and tests/presets.rs replay them against this crate. When you change bucketing or quantile behavior in either language, regenerate the fixtures (go test ./internal/ -run 'TestDatatest|TestPresets' -rewrite from the repo root) and both test suites must agree. This crate doesn't serialize snapshots, so it doesn't check the fixtures' snapshot_json line or snapshot_json.cases.

Usage

[dependencies]
goodhistogram = { git = "https://github.com/cockroachdb/goodhistogram" }

use goodhistogram::{Histogram, Params};

let h = Histogram::new(Params { lo: 1_000.0, hi: 60e9, ..Default::default() });
h.record(1_500_000); // 1.5ms in nanoseconds

let snap = h.snapshot();
let p99 = snap.value_at_quantile(0.99);

The core library has no dependencies.

Metrics-ecosystem integration (optional)

Two off-by-default features expose goodhistogram to the common Rust metrics crates. Both are duration histograms: values are observed in seconds, stored internally in nanoseconds, and exported in seconds; they export the full bucket set so histogram_quantile() works against the scraped metric.

prometheus — tikv prometheus crate

goodhistogram = { git = "https://github.com/cockroachdb/goodhistogram", features = ["prometheus"] }

use goodhistogram::{LATENCY_PARAMS, prometheus::PromHistogram};

let h = PromHistogram::new("op_duration_seconds", "op latency", LATENCY_PARAMS);
h.observe(0.0015); // seconds
registry.register(Box::new(h))?;

PromHistogram/PromHistogramVec implement prometheus::core::Collector — a near-drop-in replacement for prometheus::Histogram.

prometheus-client — OpenMetrics-native prometheus-client crate

goodhistogram = { git = "https://github.com/cockroachdb/goodhistogram", features = ["prometheus-client"] }

use goodhistogram::{LATENCY_PARAMS, prometheus_client::HistogramCollector};

let h = HistogramCollector::new("op_duration_seconds", "op latency", LATENCY_PARAMS);
h.observe(0.0015);
registry.register_collector(Box::new(h));

HistogramCollector implements prometheus_client::collector::Collector.

Planned: a metrics-facade recorder for backend-agnostic integration.