prometheus

rust-prometheus (published on crates.io as prometheus) is the Rust client library for instrumenting applications with Prometheus metrics. Its data structures and APIs are deliberately ported from the official Prometheus Go client (client_golang), giving Rust services the same Counter/Gauge/Histogram/Summary primitives, a Registry for collecting them, and both pull (/metrics scrape endpoint) and push-gateway export paths.

Maintained under the TiKV organization (the distributed key-value store behind TiDB), it’s a mature, widely used metrics library with over 127 million total downloads and a companion prometheus-static-metric crate for compile-time-optimized labeled metrics.

What You Get

• Counter, Gauge, Histogram, and their labeled *Vec variants, matching the Go client’s API shape closely
• A Registry type for collecting and exposing all registered metrics together
• Text and protobuf encoders for exposing metrics at a scrape endpoint (protobuf feature enabled by default)
• Optional process feature for standard process metrics (CPU, memory, file descriptors) on Linux via procfs
• Optional push feature for pushing metrics to a Prometheus Pushgateway
• Companion prometheus-static-metric crate (in static-metric/) that generates compile-time metric accessors for label sets known ahead of time, avoiding runtime label lookups

Common Use Cases

• Instrumenting a Rust web service or backend (as TiKV itself does) with request counters, latency histograms, and gauges
• Exposing a /metrics endpoint for Prometheus to scrape from a long-running Rust daemon
• Reducing per-request overhead of labeled metrics in hot paths via prometheus-static-metric’s compile-time-generated accessors
• Batch or short-lived job metrics reporting via the Pushgateway integration (push feature)

Under The Hood

Architecture - The crate’s ~8,400-line src/ tree separates metric type implementations (counter.rs, gauge.rs, histogram.rs, vec.rs for labeled variants), the collection/description system (desc.rs, registry.rs, metrics.rs), export encoders (encoder/ for text and protobuf formats), and optional subsystems (process_collector.rs for OS-level metrics, push.rs for Pushgateway support); a companion crate under static-metric/ uses a proc-macro to generate strongly-typed accessors for labeled metric families known at compile time, sidestepping the hash-map lookup cost of the dynamic *Vec types. Tech Stack - Depends on protobuf (default, for the binary exposition format), parking_lot for fast locking, lazy_static, fnv/memchr for hashing/parsing, and optionally reqwest (push), procfs/libc (process metrics); edition 2018, Apache-2.0 licensed, benchmarked with criterion. Code Quality - The crate has a long production track record as TiKV’s own metrics layer (a widely deployed distributed database), with examples/ demonstrating registry setup and a CHANGELOG.md tracking releases back to early versions; however GitHub activity has been low recently (no commits since December 2025), so consumers should verify current maintenance cadence before depending on it for new features. API Design - The API intentionally mirrors the Go Prometheus client’s naming and shape (Counter, Gauge, Histogram, Registry, register()), which lowers the learning curve for teams already familiar with Prometheus instrumentation in other languages, at the cost of feeling less idiomatically Rust in places (e.g. macro-heavy vector/labeled-metric construction).