feat(benchmarking): adding ERC20 benchmarking test (#3114)

* refactor: move spamoor benchmark into testify suite in test/e2e/benchmark

- Create test/e2e/benchmark/ subpackage with SpamoorSuite (testify/suite)
- Move spamoor smoke test into suite as TestSpamoorSmoke
- Split helpers into focused files: traces.go, output.go, metrics.go
- Introduce resultWriter for defer-based benchmark JSON output
- Export shared symbols from evm_test_common.go for cross-package use
- Restructure CI to fan-out benchmark jobs and fan-in publishing
- Run benchmarks on PRs only when benchmark-related files change

* fix: correct BENCH_JSON_OUTPUT path for spamoor benchmark

go test sets the working directory to the package under test, so the
env var should be relative to test/e2e/benchmark/, not test/e2e/.

* fix: place package pattern before test binary flags in benchmark CI

go test treats all arguments after an unknown flag (--evm-binary) as
test binary args, so ./benchmark/ was never recognized as a package
pattern.

* fix: adjust evm-binary path for benchmark subpackage working directory

go test sets the cwd to the package directory (test/e2e/benchmark/),
so the binary path needs an extra parent traversal.

* wip: erc20 benchmark test

* fix: exclude benchmark subpackage from make test-e2e

The benchmark package doesn't define the --binary flag that test-e2e
passes. It has its own CI workflow so it doesn't need to run here.

* fix: replace FilterLogs with header iteration and optimize spamoor config

collectBlockMetrics hit reth's 20K FilterLogs limit at high tx volumes.
Replace with direct header iteration over [startBlock, endBlock] and add
Phase 1 metrics: non-empty ratio, block interval p50/p99, gas/block and
tx/block p50/p99.

Optimize spamoor configuration for 100ms block time:
- --slot-duration 100ms, --startup-delay 0 on daemon
- throughput=50 per 100ms slot (500 tx/s per spammer)
- max_pending=50000 to avoid 3s block poll backpressure
- 5 staggered spammers with 50K txs each

Results: 55 MGas/s, 1414 TPS, 19.8% non-empty blocks (up from 6%).

* fix: improve benchmark measurement window and reliability

- Move startBlock capture after spammer creation to exclude warm-up
- Replace 20s drain sleep with smart poll (waitForDrain)
- Add deleteAllSpammers cleanup to handle stale spamoor DB entries
- Lower trace sample rate to 10% to prevent Jaeger OOM

* fix: address PR review feedback for benchmark suite

- make reth tag configurable via EV_RETH_TAG env var (default pr-140)
- fix OTLP config: remove duplicate env vars, use http/protobuf protocol
- use require.Eventually for host readiness polling
- rename requireHTTP to requireHostUp
- use non-fatal logging in resultWriter.flush deferred context
- fix stale doc comment (setupCommonEVMEnv -> SetupCommonEVMEnv)
- rename loop variable to avoid shadowing testing.TB convention
- add block/internal/executing/** to CI path trigger
- remove unused require import from output.go

* chore: specify http

* chore: filter out benchmark tests from test-e2e

* refactor: centralize reth config and lower ERC20 spammer count

move EV_RETH_TAG resolution and rpc connection limits into setupEnv
so all benchmark tests share the same reth configuration. lower ERC20
spammer count from 5 to 2 to reduce resource contention on local
hardware while keeping the loop for easy scaling on dedicated infra.

* chore: collect all traces at once

* chore: self review

* refactor: extract benchmark helpers to slim down ERC20 test body

- add blockMetricsSummary with summarize(), log(), and entries() methods
- add evNodeOverhead() for computing ProduceBlock vs ExecuteTxs overhead
- add collectTraces() suite method to deduplicate trace collection pattern
- add addEntries() convenience method on resultWriter
- slim TestERC20Throughput from ~217 to ~119 lines
- reuse collectTraces in TestSpamoorSmoke

* docs: add detailed documentation to benchmark helper methods

* ci: add ERC20 throughput benchmark job

* chore: remove span assertions

* fix: guard against drain timeout and zero-duration TPS division

- waitForDrain returns an error on timeout instead of silently logging
- guard AchievedTPS computation when steady-state duration is zero

Author: chatton

.github/workflows/benchmark.yml

@@ -77,6 +77,29 @@ jobs:
           name: spamoor-benchmark-results
           path: test/e2e/benchmark/spamoor_bench.json
 
+  # TODO: wire up to publish results once additional tests are in place.
+  erc20-benchmark:
+    name: ERC20 Throughput Benchmark
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+      - name: Set up Go
+        uses: actions/setup-go@7a3fe6cf4cb3a834922a1244abfce67bcef6a0c5 # v6.2.0
+        with:
+          go-version-file: ./go.mod
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@b5ca514318bd6ebac0fb2aedd5d36ec1b5c232a2 # v3.10.0
+      - name: Install just
+        uses: extractions/setup-just@v3
+      - name: Build binaries
+        run: just build-evm build-da
+      - name: Run ERC20 throughput test
+        run: |
+          cd test/e2e && go test -tags evm \
+            -run='^TestSpamoorSuite$/^TestERC20Throughput$' -v -timeout=15m \
+            ./benchmark/ --evm-binary=../../../build/evm
+
   # single job to push all results to gh-pages sequentially, avoiding race conditions
   publish-benchmarks:
     name: Publish Benchmark Results