[ci] Remove unaffected ports from pull request builds#21078
[ci] Remove unaffected ports from pull request builds#21078dan-shaw merged 4 commits intomicrosoft:masterfrom
Conversation
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
|
Test CI run with modified port gsl-lite available at https://dev.azure.com/vcpkg/public/_build/results?buildId=62203&view=results The tools are fetched explicitly before the parent-hashes step, to unclutter the output and facilitate output filtering. I remove the test port modification now, and then it is ready for review. |
|
Switched to |
JackBoosY
added
the
depends:vm-update
|
@JackBoosY This does not need a "vm-update". vcpkg is bootstrapped, not provisioned. |
JackBoosY
added
category:infrastructure
|
Tagged wrong label. |
| # Prefetch tools for better output | ||
| & "./vcpkg$executableExtension" fetch cmake | ||
| & "./vcpkg$executableExtension" fetch ninja | ||
| & "./vcpkg$executableExtension" fetch git |
There was a problem hiding this comment.
Technically this git and the one used on line 184 are not the same git. Our UX for fetch is... not great :)
There was a problem hiding this comment.
Yes, this is also what I thought.
What we might do on another revision is moving the fetches out of the script, next to the bootstrap step.
|
@dg0yt Thank you so much for resolving this years-long-outstanding issue with our validation pipeline! 🥳🥳🥳🥳🥳🎉🎉🎉🎉🎉🎊🎊🎊🎊🎊🥳🥳🥳🥳🥳 |
|
Initial results from the first day of this are promising. If we see this pattern for a week or so we can look at turning on #15983
(The blue and light blue are VM costs, the rest is bandwidth and storage. You can see we spend almost nothing on storage...) |
|
We still need to see some major changes to master plus pull request activity. It happens app. twice a month. For a perfect storm, merge some key script audit, and than update some pending audit PRs. |
|
Perhaps you need to track average cost per CI run, not total cost. For me, the current bottleneck is review, not CI. |
Total cost is the only thing we get yelled at for. |
Sure! But Total cost = Number of activities * Cost per activity and you cannot say this PR improved the situation after we just saw a period where the Number of activities was low. |
|
Of course! And that's why I said "looks promising" rather than "I know this will work" |
|
#21058 this will trigger a full rebuild and is a blocking issue for me right now :) |
Oops, I should have said "and than update some tiny leaf PR". Updating PRs with large impact will continue to need expensive CI runs. But anyway, it looks like PRs really do not build more than necessary for that PR after the recent merge to master. |
The other perfect storm was breaking a particular package. In the past, this would have blocked all CI builds, due to the cache miss. Hence this PR's title. |
|
great job, @dg0yt |
5 participants
What does your PR fix?
Limits the impact of the cache state of vcpkg master on pull requests. The original state leads to a waste of build resources for pull request CI runs triggered shortly after significant changes to master. Cf. Add option to remove unaffected ports from
ciaction vcpkg-tool#210.Updates the documentation for the binary caching effect of the BuildReason parameter of the test script.
Which triplets are supported/not supported? Have you updated the CI baseline?
all, no
Does your PR follow the maintainer guide?
--
If you have added/updated a port: Have you run
./vcpkg x-add-version --alland committed the result?--