Description
Is your feature request related to a problem?
Running ASV's takes a really long time, especially with old computers, and the ASVs in PRs are not run in continuous mode, which makes it hard to find regressions before a PR is merged.
pandas/.github/workflows/ci.yml
Line 78 in accf345
(master, what we run currently, asv dev)
Describe the solution you'd like
A github actions workflow that runs asv continuous -f 1.1 upstream/master HEAD
on PRs after seeing a comment such as @github-actions asv
. In order to prevent false positives, I think the bot should first merge in master to a PR before running benchmarks.
pandas/.github/workflows/comment_bot.yml
Lines 10 to 12 in accf345
(example for triggering)
Afterwards, the bot should comment the results. (I'm not sure how to do this)
I don't think it should be a check, since it takes a while to run and might be flaky. Making it a check would run it on every commit in a PR which would make CI queue times longer for everyone.
Additional context
#36860 https://pandas.pydata.org/docs/development/roadmap.html#performance-monitoring