test(fspy): benchmark tracking overhead

[wan9chi]

Motivation

fspy does not have a stable cross-platform benchmark that isolates the wall-clock cost it adds to tracked child processes. A small synthetic workload makes regressions visible without introducing real-world repository variance.

Changes

• Add one workload: four threads each open and immediately close 2,048 pre-created files.
• Measure the same dynamically linked target untracked and tracked on Linux, macOS, and Windows Namespace runners.
• Add a statically linked Linux variant to exercise seccomp user notification as well as LD_PRELOAD.
• Normalize each result as tracked/untracked overhead, then compare with the latest result from the same PR or fall back to main.
• Persist results as workflow artifacts and update a sticky PR comment, including fork PRs through a trusted workflow_run reporter.

[wan9chi]

• test(fspy): benchmark tracking overhead #566 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	cargo/​serde_json@​1.0.149 ⏵ 1.0.151
	cargo/​criterion2@​3.0.4

View full report

[github-actions]

fspy benchmark

Results for commit ecf9e9af19c581f156188679cb2a9e8d458c9ba8. Each platform uses its latest result from this PR, falling back to main.

Linux

Compared with previous PR run at 594f3ce2.

Target	Untracked	Tracked	fspy overhead	Normalized change
dynamic	15.832 ms	22.848 ms	+44.31%	-7.09%
static	19.230 ms	63.804 ms	+231.79%	-10.95%

Workload: 4 threads, 2,048 files per thread, 8,192 total open-and-close operations.

_{X64 · AMD EPYC · run 30028036239}

Macos

Compared with previous PR run at 594f3ce2.

Target	Untracked	Tracked	fspy overhead	Normalized change
dynamic	27.331 ms	31.002 ms	+13.43%	-6.92%

Workload: 4 threads, 2,048 files per thread, 8,192 total open-and-close operations.

_{ARM64 · Apple M4 Pro (Virtual) · run 30028036239}

Windows

Compared with previous PR run at 594f3ce2.

Target	Untracked	Tracked	fspy overhead	Normalized change
dynamic	75.048 ms	66.611 ms	-11.24%	-5.15%

Workload: 4 threads, 2,048 files per thread, 8,192 total open-and-close operations.

_{X64 · AMD EPYC-v4 Processor · run 30028036239}

_{This comment is updated by the Fspy Benchmark workflow.}