feat(perf): automated Pluto-vs-Charon benchmark harness + first optimization pass

.github/workflows/perf.yml

@@ -0,0 +1,81 @@
+name: Perf vs Charon
+
+on:
+  workflow_dispatch:
+    inputs:
+      tier:
+        description: "Benchmark tiers to run (1, 2, 12, or all; tier 3 needs relay connectivity)"
+        required: false
+        default: "12"
+      quick:
+        description: "Quick mode (reduced sample counts)"
+        type: boolean
+        required: false
+        default: false
+  schedule:
+    # Nightly at 03:17 UTC, tiers 1+2 only.
+    - cron: "17 3 * * *"
+
+env:
+  CHARON_VERSION: v1.7.1
+
+permissions:
+  contents: read
+
+jobs:
+  perf:
+    runs-on: ubuntu-latest
+    timeout-minutes: 90
+    steps:
+      - uses: actions/checkout@v6
+
+      - name: Checkout charon source
+        uses: actions/checkout@v6
+        with:
+          repository: ObolNetwork/charon
+          ref: ${{ env.CHARON_VERSION }}
+          path: charon
+
+      - name: Install Rust toolchain
+        uses: dtolnay/rust-toolchain@4cda84d5c5c54efe2404f9d843567869ab1699d4 # stable
+        # rust-toolchain.toml pins the actual toolchain; the action just
+        # bootstraps rustup.
+
+      - uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2
+
+      - name: Install Go
+        uses: actions/setup-go@v5
+        with:
+          go-version-file: charon/go.mod
+          cache-dependency-path: perf/go-bench/go.sum
+
+      - name: Run benchmarks
+        env:
+          TIER: ${{ inputs.tier || '12' }}
+          QUICK: ${{ inputs.quick && '1' || '' }}
+        run: |
+          # Gate on regressions once a blessed baseline is committed.
+          BASELINE=""
+          if [[ -f perf/baseline.json ]]; then
+            BASELINE="perf/baseline.json"
+          fi
+          ./perf/run.sh --tier "$TIER" ${QUICK:+--quick} \
+            ${BASELINE:+--baseline "$BASELINE"}
+
+      - name: Report to job summary
+        if: always()
+        run: |
+          if [[ -f perf/out/report.md ]]; then
+            cat perf/out/report.md >> "$GITHUB_STEP_SUMMARY"
+          fi
+
+      - name: Upload results
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: perf-results
+          path: |
+            perf/out/report.md
+            perf/out/results.json
+            perf/out/go-bench.txt
+          if-no-files-found: warn

.gitignore

@@ -32,4 +32,6 @@ test-infra/sszfixtures/sszfixtures
 
 .claude/worktrees/
 .claude/scheduled_tasks.lock
-test-cluster
+test-cluster
+# Performance harness artifacts
+/perf/out/

Cargo.toml

@@ -176,6 +176,12 @@ panicking_overflow_checks = "deny"
 redundant_test_prefix = "deny"
 unwrap_used = "deny"
 
+# Thin LTO + a single codegen unit measured ~6-12% faster on pure-Rust hot
+# paths (SSZ, k1util) with no effect on blst-bound ones; see perf/FINDINGS.md.
+[profile.release]
+lto = "thin"
+codegen-units = 1
+
 # Optimize crypto dependencies in test builds for faster tests
 [profile.test.package.scrypt]
 opt-level = 3

crates/core/Cargo.toml

@@ -6,6 +6,14 @@ repository.workspace = true
 license.workspace = true
 publish.workspace = true
 
+[lib]
+bench = false
+
+[features]
+# Exposes internal modules and in-memory harnesses to the criterion benches.
+# Never enable in production builds.
+bench-util = []
+
 [dependencies]
 alloy.workspace = true
 backon.workspace = true
@@ -44,6 +52,7 @@ url.workspace = true
 [dev-dependencies]
 anyhow.workspace = true
 clap.workspace = true
+criterion.workspace = true
 rand.workspace = true
 prost.workspace = true
 prost-types.workspace = true
@@ -58,6 +67,21 @@ tokio = { workspace = true, features = ["test-util"] }
 wiremock.workspace = true
 tower = { workspace = true, features = ["util"] }
 
+[[bench]]
+name = "ssz_codec"
+harness = false
+path = "benches/ssz_codec.rs"
+
+[[bench]]
+name = "corepb"
+harness = false
+path = "benches/corepb.rs"
+
+[[bench]]
+name = "qbft"
+harness = false
+path = "benches/qbft.rs"
+
 [build-dependencies]
 pluto-build-proto.workspace = true
 built.workspace = true

crates/core/benches/corepb.rs

@@ -0,0 +1,45 @@
+//! Benchmarks for protobuf (de)serialization of consensus wire messages.
+//!
+//! Benchmark ids follow the cross-language pair naming used by
+//! `perf/pairs.json`; the Go counterparts live in `perf/go-bench/`. The
+//! fixture is a round-3 QBFTConsensusMsg with four justifications and a 1KiB
+//! value, generated by the Go side.
+#![allow(missing_docs)]
+
+use std::{hint::black_box, time::Duration};
+
+use criterion::{Criterion, criterion_group, criterion_main};
+use pluto_core::corepb::v1::consensus::QbftConsensusMsg;
+use prost::Message;
+
+const QBFT_MSG_PB: &[u8] = include_bytes!("../../../perf/fixtures/qbft_consensus_msg.pb");
+
+fn bench_qbft_msg(c: &mut Criterion) {
+    let msg = QbftConsensusMsg::decode(QBFT_MSG_PB).expect("bench setup: decode fixture");
+    assert_eq!(
+        msg.encode_to_vec(),
+        QBFT_MSG_PB,
+        "qbft msg fixture round-trip mismatch"
+    );
+
+    c.bench_function("tier1/proto/qbft_marshal", |b| {
+        b.iter(|| black_box(&msg).encode_to_vec())
+    });
+
+    c.bench_function("tier1/proto/qbft_unmarshal", |b| {
+        b.iter(|| QbftConsensusMsg::decode(black_box(QBFT_MSG_PB)).expect("decode should succeed"))
+    });
+}
+
+fn config() -> Criterion {
+    Criterion::default()
+        .sample_size(50)
+        .measurement_time(Duration::from_secs(3))
+}
+
+criterion_group! {
+    name = benches;
+    config = config();
+    targets = bench_qbft_msg,
+}
+criterion_main!(benches);