Skip to content

README.md

Latest commit

 

History

History
57 lines (38 loc) · 1.55 KB

README.md

File metadata and controls

57 lines (38 loc) · 1.55 KB

Profiling

This directory contains reproducible profiling scripts for indicators-cli.

Quick Start (Via CLI)

Run profiling directly from the command line without separate scripts:

# Run all profiling suites in cached mode (fast, deterministic)
indicators --profile --profile-mode cached

# Run only e2e profiling
indicators --profile --profile-suite e2e --profile-mode cached

# Run one specific scenario  
indicators --profile --profile-scenario e2e_cpu_cached

Scripts and Direct Use

For advanced workflows, you can also run scripts directly:

python profiling/run_profiles.py --suite all --mode cached
python profiling/profile_e2e.py --scenario e2e_cpu_cached

Scenarios

Default scenarios are defined in profiling/scenarios.json.

Key fields:

  • suite: e2e, source, calculation, or write
  • mode: live or cached
  • engine: cpu or gpu
  • repeat: number of iterations

Live Mode Prerequisite

Live source and e2e scenarios go through the current source_data conversion path, which requires pyarrow in this environment.

pip install pyarrow

Outputs

Outputs are written to profiling/results/:

  • Per-suite JSON summary: profiling/results/<suite>_<run_id>.json
  • Per-suite CSV rows: profiling/results/<suite>_<run_id>.csv
  • Aggregate JSON: profiling/results/latest.json
  • Aggregate CSV: profiling/results/latest.csv
  • Raw cProfile files: profiling/results/raw/*.prof

GPU scenarios are auto-skipped when GPU execution is unavailable, and skip reasons are included in JSON/CSV artifacts.