Use Depot hosted runners instead of GHA hosted runners. - #22993
Conversation
Depot have reached out about this, so I want to see how easy it is.
benjyw
added
the
release-notes:not-required
|
Nice! The Depot Linux x86 shards seem to be quite a bit faster than on GHA hardware. But I need to do a fully uncached build on both to verify apples-to-apples. |
|
👀 oh cool, is that a REAPI offering too? https://depot.dev/products/cache |
|
Oh damn, Depot coming in hot... A Pants client used them, and they were wayyy faster |
|
We now have access to their macos runners, so this should all pass. Once this is in I will look at switching to them for Linux ARM as well, and to trying out their REAPI offering. |
|
Is there a rough estimate on how much faster these are? The CI is tricky, since it sporadically uses cache. Also, for running CI in forks, this would/should fail out of the box right? This is my ignorance related to how Depot was setup on the backend (e.g. is it whitelisted? Would this apply to scie-pants too?) |
This PR was on a fork and it worked... |
It was a PR from your fork to pantsbuild/pants right? I'm referring to running CI in the fork itself. |
|
Ah, that would require a separate depot account for that fork I think. So perhaps this should be switchable? |
Nahhh, no part of the Pants offering or documentation is making GHA in other forks work out of the box. It doesn't today anyways. I was just checking, as I have a patch that I can apply (or we can make the generate script handle this). Not important as part of this PR. |
|
🚀 ! Excited to see these in action. |
…2993) Depot have kindly offered use of their runners as in-kind sponsorship.
4 participants
Depot have reached out about this, so I want to see how easy it is.