Skip to content

SOLDR_BUILD_PERF.md

Latest commit

 

History

History
50 lines (38 loc) · 1.63 KB

SOLDR_BUILD_PERF.md

File metadata and controls

50 lines (38 loc) · 1.63 KB

soldr Rust Build Performance

Measured local soldr performance for building the Rust fbuild CLI and daemon from this workspace.

Environment

  • Host: Windows, x86_64-pc-windows-msvc
  • Tool wrapper: soldr 0.7.4
  • Managed cache: zccache 1.3.0
  • Rust toolchain: rustc 1.94.1
  • Command:
uv run soldr cargo build --release -p fbuild-cli -p fbuild-daemon

The benchmark used temporary target directories under .cache/ so the normal workspace target/ directory was not part of the measurement.

Results

Case Time
Cold: empty target dir and cleared soldr cache 165.536 s
Warm cache: same target dir after cargo clean, hot soldr cache 55.898 s
No-op rebuild: same target dir, no source changes 0.787 s
Fresh different target dir with hot soldr cache 162.834 s

Notes

soldr is correctly routing Cargo through its managed zccache, but cache reuse was target-directory sensitive in this local run. Rebuilding the same target directory after cargo clean was about 3x faster than the cold build, while compiling into a different fresh target directory got effectively no benefit.

Final cache status after the run reported:

Compilations: 1393 total (216 cached, 849 cold, 311 non-cacheable)
Hit rate:     20.3%
Time saved:   ~9m 21s
Artifacts:    849 (2.5 GB)

An initial stale cache-daemon state was observed before the benchmark: soldr cargo --version reported that a daemon PID existed but was not accepting connections. Running soldr cargo --version again started a healthy managed zccache daemon, after which the timed runs completed successfully.