relabel: add Prometheus-compatible target and metric relabeling

AGENTS.md

@@ -24,7 +24,7 @@ Do not use Rust language or std features newer than `rust-version` in `Cargo.tom
 
 ## Spec
 
-[specs/expfmt/SPEC.md](specs/expfmt/SPEC.md) is the contract for `expfmt`. If you change behavior the spec describes, update the spec in the same PR.
+[specs/SPEC.md](specs/SPEC.md) is the specification contract for `scrape.rs` (`expfmt`, `relabel`, and scrape pipeline integration). If you change behavior the spec describes, update the spec in the same PR.
 
 ## Test data
 

Cargo.toml

@@ -13,28 +13,37 @@ categories = ["parser-implementations", "encoding"]
 exclude = ["tests/otap-compat"]
 
 [features]
-default = ["arrow", "dto"]
+default = ["arrow", "dto", "relabel"]
 arrow = ["dep:arrow-array", "dep:arrow-buffer", "dep:arrow-schema"]
 dto = ["dep:prost", "dep:prost-types"]
+relabel = ["dep:regex", "dep:serde"]
 
 [dependencies]
 arrow-array = { version = "59", default-features = false, optional = true }
 arrow-buffer = { version = "59", default-features = false, optional = true }
 arrow-schema = { version = "59", default-features = false, optional = true }
 prost = { version = "0.14", optional = true }
 prost-types = { version = "0.14", optional = true }
+regex = { version = "1", optional = true }
+serde = { version = "1", features = ["derive"], optional = true }
 thiserror = "2"
 
 [dev-dependencies]
 benchfmt = { git = "https://github.com/bwplotka/benchfmt.rs" }
 prost-build = "=0.14.4"
 protox = "0.10"
+serde_json = "1"
 
 [[bench]]
 name = "decode"
 harness = false
 required-features = ["arrow"]
 
+[[bench]]
+name = "relabel"
+harness = false
+required-features = ["relabel"]
+
 [lints.rust]
 missing_docs = "warn"
 unsafe_code = "forbid"

README.md

@@ -13,21 +13,23 @@ Service discoveries
   -> Target relabeling
   -> Scrape manager / HTTP fetch
   -> Streaming expfmt parser
-  -> Metric relabeling / drop hook
+  -> Stage 1: family `__name__` filter pushdown
+  -> Metric family assembly & string interning
+  -> Stage 2: dictionary-vectorized relabeling & `honor_labels`
   -> Columnar Apache Arrow (OTAP) batches or Protobuf DTOs
 ```
 
 The project serves two main use cases:
 
-1. Columnar scraping for OpenTelemetry Arrow. [`open-telemetry/otel-arrow#3277`](https://github.com/open-telemetry/otel-arrow/issues/3277) defines an OTAP-native `prometheus_scrape` receiver in Rust. `scrape.rs` streams scrape bytes through pre-allocation relabeling and metric family assembly directly into Apache Arrow (`OtapArrowRecords`) tables on thread-per-core `!Send` runtimes with bounded memory.
+1. Columnar scraping for OpenTelemetry Arrow. [`open-telemetry/otel-arrow#3277`](https://github.com/open-telemetry/otel-arrow/issues/3277) defines an OTAP-native `prometheus_scrape` receiver in Rust. `scrape.rs` streams scrape bytes through early `__name__` family filtering, metric family assembly, and dictionary-vectorized relabeling directly into Apache Arrow (`OtapArrowRecords`) tables on thread-per-core `!Send` runtimes with bounded memory.
 2. Standalone Rust Prometheus tooling. Validation harnesses such as OpenTelemetry Weaver live-checking `/metrics` endpoints against semantic conventions, CLIs, and custom collectors can use individual components (such as `expfmt` DTO decoding or `relabel`) without pulling in the full scrape runtime.
 
 ## Components and scope
 
 | Component | Status | Specification | Scope |
 | :--- | :--- | :--- | :--- |
-| `expfmt` | In progress | [specs/expfmt/SPEC.md](./specs/expfmt/SPEC.md) | `Content-Type` and `Accept` negotiation, zero-copy streaming parsers for Prometheus Text 0.0.4, OpenMetrics 1.0, OpenMetrics 2.0, and Prometheus Protobuf, primary Apache Arrow / OTAP decoder (`ArrowDecoder`), and [`MetricFamily`](https://github.com/prometheus/prometheus/blob/main/prompb/io/prometheus/client/metrics.proto) protobuf DTO decoder. |
-| `relabel` | Planned | TBD | Prometheus-compatible target and metric relabeling (`replace`, `keep`, `drop`, `hashmod`, `labelmap`, `labeldrop`, `labelkeep`, `lowercase`, `uppercase`, `keepequal`, `dropequal`), including pre-allocation filtering on borrowed parser label buffers. |
+| `expfmt` | In progress | [specs/SPEC.md](./specs/SPEC.md) | `Content-Type` and `Accept` negotiation, zero-copy streaming parsers for Prometheus Text 0.0.4, OpenMetrics 1.0, OpenMetrics 2.0, and Prometheus Protobuf, primary Apache Arrow / OTAP decoder (`ArrowDecoder`), and [`MetricFamily`](https://github.com/prometheus/prometheus/blob/main/prompb/io/prometheus/client/metrics.proto) protobuf DTO decoder. |
+| `relabel` | In progress | [specs/SPEC.md](./specs/SPEC.md#target-and-metric-relabeling-architecture) | Prometheus-compatible target and metric relabeling (`replace`, `keep`, `drop`, `hashmod`, `labelmap`, `labeldrop`, `labelkeep`, `lowercase`, `uppercase`, `keepequal`, `dropequal`), Stage 1 `__name__` filter pushdown, and Stage 2 dictionary-vectorized relabeling and `honor_labels` resolution. |
 | `discovery` | Planned | TBD | Standard Prometheus service discovery mechanisms (static targets, file-based SD, HTTP SD, DNS SD, Kubernetes SD) emitting target groups and `__meta_*` labels. |
 | `scrape` | Planned | TBD | Bounded HTTP scrape scheduling, target pool lifecycle, compression and content negotiation, stale marker handling, target state reporting, and self-telemetry. |
 
@@ -51,8 +53,9 @@ Run the Go differential fixture extractor and tests (requires Go):
 cd tests/go-harness && go test ./...
 ```
 
-Run the decoding benchmarks in Go `benchfmt` format for [`benchstat`](https://pkg.go.dev/golang.org/x/perf/cmd/benchstat):
+Run the decoding and relabeling benchmarks in Go `benchfmt` format for [`benchstat`](https://pkg.go.dev/golang.org/x/perf/cmd/benchstat):
 
 ```sh
 cargo bench --bench decode -- -bench '^BenchmarkDecode' -benchtime 1s -count 6 -cpu 2 -benchmem | tee decode.txt
+cargo bench --bench relabel -- -bench '^BenchmarkRelabel' -benchtime 1s -count 6 -cpu 2 -benchmem | tee relabel.txt
 ```