[joseph-isaacs]

Summary

After #10987, nothing reads s3://vortex-ci-benchmark-results/data.json.gz or commits.json. Appending to the 353 MB data.json.gz (download, gunzip, append, gzip, upload, with ETag retries across parallel jobs) was the slowest step of every develop benchmark job. All results are already ingested into the benchmarks Postgres.

[codspeed]

Merging this PR will improve performance by 10.24%

1 warning

⚠️ Walltime results may be unreliable

Some benchmarks used the Walltime instrument outside CodSpeed Macro Runners, which can make the results inconsistent.

Learn more about Macro Runners

See affected benchmarks

⚡ 1 improved benchmark
✅ 2317 untouched benchmarks

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Walltime	threshold_avx2[80]	3 µs	2.7 µs	+10.24%

Tip

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

---

_{Comparing ji/bench-stop-s3-results (3e8ffa7) with develop (05fd248)}

[joseph-isaacs]

The CodSpeed failure on 66141e3d (bitpack_compare_sweep compare[1] −15.9%, starts_with_avx512[64, VarBin] −14.9%, alongside a +18% improvement elsewhere) is benchmark noise, not this PR. This stack only changes GitHub workflows and Python scripts under scripts/; no Rust code changed, so nothing CodSpeed measures can move. The same kind of swing showed up on #10987 earlier and cleared on the next head. A newer head has already been pushed, so CodSpeed will run again there.

---

Generated by Claude Code

[joseph-isaacs]

CUDA build & lint failed on a20275f3 before any step ran. The runner couldn't download its setup action: Unable to resolve action runs-on/action@v2, unable to find version v2. That's a transient GitHub/runner problem, not this PR. A newer head has been pushed since, so the job runs again there.

---

Generated by Claude Code