Skip to content

Latest commit

 

History

History

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 

README.md

Results

This directory holds published benchmark measurements. No number appears here unless it came from an actual CI run — see ../methodology.md for the rules (single variable runs-on:, shape parity, median of at least three runs, cold vs warm cache reported separately, upstream project pinned to a specific commit).

How results are published

  1. Both caller workflows for a case study are run at least three times each, for each cache state (cold and warm), using workflow_dispatch.

  2. Each run's job wall-clock is read from the GitHub Actions jobs API (started_at .. completed_at), excluding run orchestration and queue time.

  3. The runs are reduced to a median per (case study, runner, cache state):

    python scripts/case_study_collect.py \
        --repo tempusbuild/benchmark \
        --upstream django/django \
        --ref <upstream-commit-sha> \
        --tempus-workflow casestudy-django-tempus.yml \
        --github-workflow casestudy-django-github.yml

    This prints the Markdown rows that replace the table below.

  4. The run count, upstream ref, cache state, and date are recorded with every row so a reader can judge and reproduce the number.

Latest results

The table below is the template. It is intentionally empty — it will be populated only with measured medians, never estimates. There is one row per (upstream project, runner, cache state).

Upstream Ref Runner Runs Median wall-clock Cache Date
django/django ubuntu-latest cold
django/django tempus-ubuntu-24.04-4core cold

(Warm-cache rows are added the same way, with warm in the Cache column. Cold and warm are never blended in a single row.)

Column meanings:

  • Upstream — the real OSS project whose own test command was run.
  • Ref — the pinned upstream commit SHA the runs measured (first 7 chars shown).
  • Runner — the runs-on: label.
  • Runs — number of runs the median is taken over (at least three).
  • Median wall-clock — median total measured wall-clock time across those runs.
  • Cachecold (no primed dependency cache) or warm (primed). Never blended.
  • Date — date the runs were collected.