You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@airflow.apache.org by "zozo123 (via GitHub)" <gi...@apache.org> on 2026/10/05 10:05:00 UTC
Re: [PR] Accelerate CI image materialization and dependency reuse [airflow]
zozo123 commented on PR #74173:
URL: https://github.com/apache/airflow/pull/74173#issuecomment-5992299811
Indicative only: one treatment run and 1 exact-matrix baseline runs. Consumer preparation changed by -83.0 seconds, all-job workflow wall time by -1.5 minutes, and observed runner time by -69.6 minutes. These observational differences do not establish causation or a weekly saving.
| Metric | Baseline median (p90) | PR median (p90) | Delta PR − baseline | n baseline / PR |
|---|---:|---:|---:|---:|
| Prep seconds | 210.0 (250.2) | 127.0 (174.2) | -83.0 | 79 / 79 |
| Consumer gap after builder seconds | 3.0 (4.0) | 3.0 (4.0) | +0.0 | 79 / 79 |
| Consumer start from workflow seconds | 583.0 (584.0) | 652.0 (653.0) | +69.0 | 79 / 79 |
| Workflow wall minutes (all executed jobs) | 51.2 (51.2) | 49.8 (49.8) | -1.5 | 1 / 1 |
| Runner minutes (all executed jobs) | 1306.0 (1306.0) | 1236.4 (1236.4) | -69.6 | 1 / 1 |
| Builder duration seconds | 486.0 (486.0) | 541.0 (541.0) | +55.0 | 1 / 1 |
| Snapshot create+upload seconds | 0.0 (0.0) | 96.0 (96.0) | +96.0 | 1 / 1 |
| Initial dispatch seconds | 65.0 (65.0) | 84.0 (84.0) | +19.0 | 1 / 1 |
Read-only timing analysis at f53486a4. Snapshot creation/upload: 73s + 23s; restore outcomes: {'restored': 79}. One treatment attempt is insufficient to claim ROI. Baselines match active job names and exclude CI-related file changes; wall time covers this AMD workflow, not verified required checks. True per-job queue time and dependency-cache benefit remain unmeasured. Please collect 2–4 additional full attempts and additional matched baselines before drawing conclusions.
Source runs: [treatment, attempt 1](https://github.com/apache/airflow/actions/runs/37236836509) and [exact-matrix baseline, attempt 1](https://github.com/apache/airflow/actions/runs/37267385329). The 79 prep/start observations in each group are jobs within a single run, not independent workflow samples. p90 uses linear interpolation. Runner-minutes are observed execution time, not billed cost.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@airflow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org