perf(assertion-executor): perf-only hot-path wins and profiling loop by odyslam · Pull Request #675 · phylaxsystems/credible-sdk

odyslam · 2026-03-14T10:46:50Z

Summary

lands only the perf-path changes extracted from the stacked executor PRs; no new cheatcodes or other business-logic feature work
keeps the executor, tracer, store, and precompile hot-path optimizations together with the benchmark/profiling scaffolding used to isolate and validate them
hardens the new perf paths with targeted tests, CI cleanup, and explanatory comments for reviewers

Review Guide

Executor hot path
- crates/assertion-executor/src/executor/mod.rs
- crates/assertion-executor/src/executor/with_inspector.rs
- empty-selector short-circuits, single-selector direct execution, and small-batch scheduling
Tracer / store hot path
- crates/assertion-executor/src/inspectors/tracer.rs
- crates/assertion-executor/src/store/assertion_store.rs
- crates/assertion-executor/src/inspectors/precompiles/legacy/get_logs.rs
- crates/assertion-executor/src/inspectors/precompiles/legacy/state_changes.rs
- cached log encoding, tx-log indexing, storage/write indexing, and lower-allocation trigger reads
Benchmarking / profiling loop
- crates/benchmark-utils/src/perf.rs
- crates/benchmark-utils/src/bin/benchmark-utils-perf.rs
- crates/assertion-executor/benches/
- Makefile
- .github/workflows/benchmarks.yml
- .claude/skills/samply-profile/SKILL.md
- phase-isolated Criterion benches plus the warmed samply entrypoint

Non-goals

no new cheatcodes
no new trigger semantics or assertion feature work
no intended behavior changes beyond preserving the existing assertion path while reducing setup and tracing overhead

Validation

cargo clippy -p assertion-executor -p benchmark-utils --all-targets --features test -- -D warnings -D clippy::pedantic
cargo test -p benchmark-utils
cargo test -p assertion-executor --lib
cargo bench --manifest-path crates/assertion-executor/Cargo.toml --features test --benches --no-run
RUSTFLAGS='-C opt-level=1' SIDECAR_BENCH_TRANSPORT=mock RUST_LOG=error cargo bench -p sidecar --features bench-utils --benches --no-run

Representative Bench Deltas Vs main

assertion_store::read/miss_nonexistent_assertion: about -97.7%
call_tracer_truncate/15k_deep_pending: about -19.4%
avg_block_100_aa: about -32.5%
erc20_transaction: about -65.5%
uniswap_transaction_aa: about -28.1%

Samply Follow-up

added warmed samply runs via benchmark-utils-perf
the warmed profiles currently point to assertion-side account-load / db-clone setup churn as the next likely optimization target

github-actions · 2026-03-14T11:32:39Z

Benchmark report is ready.

Report: https://phylaxsystems.github.io/credible-sdk/main-bench/pr-675/report/
No significant benchmark changes detected.

github-actions · 2026-03-14T12:32:30Z

Benchmark report is ready.

Report: https://phylaxsystems.github.io/credible-sdk/main-bench/pr-675/report/
No significant benchmark changes detected.

…path Assertions that match no trigger selectors were previously returned from the store read path with an empty selector set, causing unnecessary setup work (persistent account insertion, MultiForkDb creation) in the executor. Changed read_adopter from .map() to .filter_map() so assertions with zero matched selectors are dropped before reaching execution. Tests: assertion_store suite (32/32 pass), full lib suite (210 pass, 1 pre-existing failure from missing test artifact) Bench baseline (before change): assertion_store::read/hit_existing_assertion: 679.68 ns assertion_store::read/miss_nonexistent: 324.00 ns executor_transaction_performance/eoa_transaction_aa: 44.824 µs executor_transaction_performance/erc20_transaction_aa: 67.632 µs executor_transaction_performance/uniswap_transaction_aa: 387.06 µs Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> (cherry picked from commit a10e259)

…xecution Both run_assertion_contract and run_assertion_contract_with_inspector now return immediately when fn_selectors is empty, avoiding unnecessary prepare_assertion_contract work (persistent account insertion, MultiForkDb creation, PhEvmInspector cloning). Combined with Phase 1 selector-pruning, this ensures zero-work execution when trigger matching produces no selectors. Tests: executor suite (6/6 pass) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> (cherry picked from commit dac36b2)

Adds a PARALLEL_THRESHOLD constant (currently 2). When the number of assertion contracts or fn_selectors is below this threshold, execution uses sequential iterators instead of rayon's parallel iterators to avoid thread scheduling overhead. This matters because: - Most transactions trigger 0-1 assertion contracts - Most assertion contracts have 1-3 fn_selectors - Rayon fan-out has ~10-20µs scheduling overhead per task The change is behavior-preserving: only scheduling strategy changes. Also re-exports AssertionsForExecution from store module for use in the executor's sequential path closure type annotation. Tests: full lib suite (210 pass, 1 pre-existing failure) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> (cherry picked from commit 1d72940)

github-actions · 2026-03-14T15:05:31Z

Benchmark report is ready.

Report: https://phylaxsystems.github.io/credible-sdk/main-bench/pr-675/report/
No significant benchmark changes detected.

github-actions · 2026-03-14T15:41:19Z

Benchmark report is ready.

Report: https://phylaxsystems.github.io/credible-sdk/main-bench/pr-675/report/
No significant benchmark changes detected.

github-actions · 2026-03-14T18:23:29Z

Benchmark report is ready.

Report: https://phylaxsystems.github.io/credible-sdk/main-bench/pr-675/report/
No significant benchmark changes detected.

odyslam · 2026-03-17T02:06:50Z

Updated the PR title/body to make review easier.

Suggested review order:

executor hot path (executor/mod.rs, executor/with_inspector.rs)
tracer/store hot path (inspectors/tracer.rs, store/assertion_store.rs, legacy precompile caches)
benchmarking/profiling scaffolding (benchmark-utils, benches, Makefile, workflow, samply skill)

This PR is intentionally perf-only: no new cheatcodes and no new business-logic feature work.

crates/assertion-executor/src/store/mod.rs

.claude/skills/samply-profile/SKILL.md

github-actions · 2026-03-19T16:11:24Z

Benchmark report is ready.

Report: https://phylaxsystems.github.io/credible-sdk/main-bench/pr-675/report/
No significant benchmark changes detected.

odyslam temporarily deployed to github-pages March 14, 2026 11:32 — with GitHub Actions Inactive

odyslam changed the title ~~Fix assertion_store hot path regression~~ Bucket executor performance improvements Mar 14, 2026

odyslam changed the base branch from stack/06-final-polish to main March 14, 2026 12:16

odyslam temporarily deployed to github-pages March 14, 2026 12:32 — with GitHub Actions Inactive

odyslam force-pushed the codex/final-polish-bench branch from 696078b to f0d5516 Compare March 14, 2026 13:13

odyslam changed the title ~~Bucket executor performance improvements~~ perf(assertion-executor): extract perf-only executor and tracer wins Mar 14, 2026

odyslam and others added 7 commits March 14, 2026 15:29

perf(assertion-executor): bucket perf-only executor optimizations

7ad12ee

perf(assertion-executor): keep tracer changes cache-only

4daccd7

perf(assertion-executor): extract executor and tracer hot-path wins

69307fe

fix(assertion-executor): repair rebased perf branch

d99f7d6

odyslam force-pushed the codex/final-polish-bench branch from 6db8e8a to d99f7d6 Compare March 14, 2026 14:40

odyslam temporarily deployed to github-pages March 14, 2026 15:05 — with GitHub Actions Inactive

test(assertion-executor): harden perf path regressions

09cf4da

odyslam temporarily deployed to github-pages March 14, 2026 15:40 — with GitHub Actions Inactive

docs(assertion-executor): clarify perf-path control flow

f9b854d

odyslam temporarily deployed to github-pages March 14, 2026 18:23 — with GitHub Actions Inactive

odyslam added 2 commits March 14, 2026 22:44

Comment perf profiling loop changes

b7af418

Fix CI for perf loop changes

fb85ef3

odyslam changed the title ~~perf(assertion-executor): extract perf-only executor and tracer wins~~ perf(assertion-executor): perf-only hot-path wins and profiling loop Mar 17, 2026

odyslam marked this pull request as ready for review March 17, 2026 02:10

odyslam requested a review from makemake-kbo as a code owner March 17, 2026 02:10

odyslam requested review from fredo, lean-apple and mateo-mro as code owners March 17, 2026 02:10

claude bot reviewed Mar 17, 2026

View reviewed changes

crates/assertion-executor/src/store/mod.rs Outdated Show resolved Hide resolved

.claude/skills/samply-profile/SKILL.md Show resolved Hide resolved

odyslam added 4 commits March 16, 2026 22:54

Address PR review comments

4e992ea

Fix benchmark-only rayon import

2271c0f

fix ci failures on benchmark and advisories

74f575d

fix rustdoc warnings in docs jobs

7390a53

odyslam deployed to github-pages March 19, 2026 16:11 — with GitHub Actions View deployment

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

perf(assertion-executor): perf-only hot-path wins and profiling loop#675

perf(assertion-executor): perf-only hot-path wins and profiling loop#675
odyslam wants to merge 15 commits intomainfrom
codex/final-polish-bench

odyslam commented Mar 14, 2026 •

edited

Loading

Uh oh!

github-actions bot commented Mar 14, 2026

Uh oh!

github-actions bot commented Mar 14, 2026

Uh oh!

github-actions bot commented Mar 14, 2026

Uh oh!

github-actions bot commented Mar 14, 2026

Uh oh!

github-actions bot commented Mar 14, 2026

Uh oh!

odyslam commented Mar 17, 2026

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Mar 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

odyslam commented Mar 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Review Guide

Non-goals

Validation

Representative Bench Deltas Vs main

Samply Follow-up

Uh oh!

github-actions bot commented Mar 14, 2026

Uh oh!

github-actions bot commented Mar 14, 2026

Uh oh!

github-actions bot commented Mar 14, 2026

Uh oh!

github-actions bot commented Mar 14, 2026

Uh oh!

github-actions bot commented Mar 14, 2026

Uh oh!

odyslam commented Mar 17, 2026

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Mar 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

odyslam commented Mar 14, 2026 •

edited

Loading