perf(memtrack): reduce benchmark noise and resolve workload symbols

[not-matthias]

Changes

• Remove the short ls benchmark, whose measured time and variation are dominated by allocator-uprobe teardown.
• Increase tar warmup from 5s to 30s while retaining real archive output and the 60s measurement budget.
• Install exact-version, architecture-matched tar/coreutils debug packages from Launchpad over HTTPS before profiling. Do not upgrade workload binaries or depend on mutable ddebs indexes.

Missing symbols

The Ubuntu workload binaries lack full symbols, and the pinned Samply fork explicitly blocks Ubuntu debuginfod. Existing memtrack Rust names were already present. Installing local matching debug files resolves workload names and source locations before presymbolication.

All 20 hosted flamegraphs from the five executions have named tar/dd workload nodes. Tar graphs contain 181–212 workload nodes each, all named; dd graphs contain 23–35, all named. Tar's named hot frames consistently include flush_archive and sys_write_archive_buffer/flush_write. dd's leading named costs consistently include BPF CO-RE candidate matching and BTF initialization.

High-address frames with no mapped object remain unresolved (18.5–21.6% of sampled self-CPU weight in tar). Kernel-side attribution is plausible, but its cause is not established by these graphs. This change does not claim to resolve those frames.

Five-run validation

Exact revision: a2c9b21. All five tar and dd benchmark jobs passed. Four complete workflows passed; one failed only because macOS clippy could not resolve index.crates.io.

Benchmark	Range (seconds)	Sample CV
tar	9.299–9.403	0.45%
tar, physical	9.427–9.497	0.30%
dd	1.139–1.251	3.67%
dd, physical	1.228–1.326	2.92%

Each tar variant completed four warmup rounds and seven measured rounds. The first measured workload span was within 0.65% of the following-round median in every execution; event counts stayed at 24,902 normally and 26,403 with physical tracking. These five runs show low tar variation, not a guarantee of future stability. dd remains noisier.

CI executions: 36602936493, 36602947014, 36602957193, 36602966329, 36602976080.

• Relevant prek hooks passed.
• Ubuntu 22.04 container smoke verified unchanged tar/dd executable checksums and build-ID-matched debug files containing symbols and DWARF.
• An earlier debug-package approach hit inconsistent Ubuntu ARM64 mirror indexes and implicitly upgraded workload packages. Those runs are excluded from the statistics above; exact-version Launchpad downloads avoid both problems.

[codspeed]

Merging this PR will improve performance by 11.7%

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚡ 1 improved benchmark
✅ 30 untouched benchmarks
⏩ 6 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	WallTime	memtrack track tar	9.4 s	8.4 s	+11.7%

Tip

Curious why performance improved? Comment @codspeedbot explain why performance improved on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing cod-3690-investigate-flaky-ls-and-tar-benchmarks (407b28e) with main (c01f5d2)}

Footnotes

1. 6 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[greptile-apps]

Confidence Score: 5/5

[Low risk] Adjusts benchmark configuration and removes a benchmark workload.

The PR appears safe to merge; no outstanding or new actionable finding remains.

Summary

The PR removes the teardown-dominated ls benchmark and increases tar warmup from 5 to 30 seconds while retaining the 60-second measurement budget.

• CI continues to run the dd and tar workload configurations.
• No new actionable issue was identified.

_{Reviews (3) · Last reviewed commit: "perf(memtrack): warm up tar for several ..."}