Problem
A push to main that touches a benchmark-relevant file starts a full benchmark run, and the deploy job can publish it to gh-pages. During the performance incident, a push caused a run that had to be cancelled mid-production deployment to avoid overwriting the day’s data with a contaminated point.
PR #143 prevents partial manual dispatches from deploying, but push-triggered full runs still pass the current deploy condition.
Proposed policy
Only scheduled runs and explicit, non-partial manual dispatches should publish benchmark history. Push runs should retain logs/artifacts for validation but should not overwrite the canonical daily data.
Also review whether every scripts/ change needs to trigger the benchmark matrix; processing-only changes may not.
Acceptance criteria
Source
Performance investigation: item 5 — push-to-main deploys gh-pages
Problem
A push to
mainthat touches a benchmark-relevant file starts a full benchmark run, and the deploy job can publish it to gh-pages. During the performance incident, a push caused a run that had to be cancelled mid-production deployment to avoid overwriting the day’s data with a contaminated point.PR #143 prevents partial manual dispatches from deploying, but push-triggered full runs still pass the current deploy condition.
Proposed policy
Only scheduled runs and explicit, non-partial manual dispatches should publish benchmark history. Push runs should retain logs/artifacts for validation but should not overwrite the canonical daily data.
Also review whether every
scripts/change needs to trigger the benchmark matrix; processing-only changes may not.Acceptance criteria
Source
Performance investigation: item 5 — push-to-main deploys gh-pages