ray icon indicating copy to clipboard operation
ray copied to clipboard

[release-automation] Separate steps into groups

Open khluu opened this issue 1 year ago • 6 comments

  • Apparently the block step blocks all the steps after it (unless the step specifies depends_on). This happens for mac sanity check step since it has no dependency: https://buildkite.com/ray-project/release-automation/builds/470/dag
  • This separates each platform sanity check to separate groups so they don't block each other
  • Also separate upload wheels step into its own group to allow more flexibility (upload to PyPI & TestPyPI doesn't have to be in the same build)
  • Fix typo forge-arm64 -> forge_arm64

khluu avatar May 15 '24 09:05 khluu

I tested with this build and was able to see each of the group start "waiting for agent" right after unblocking without waiting for other steps: https://buildkite.com/ray-project/release-automation/builds/476#_

khluu avatar May 15 '24 09:05 khluu

isn't it easier to just add the depends_on for steps? rather than breaking it down into pieces?

aslonnie avatar May 15 '24 15:05 aslonnie

isn't it easier to just add the depends_on for steps? rather than breaking it down into pieces?

for example, mac sanity checks don't really have any depends_on. I tried having it depends on the block step but that didn't really work.

khluu avatar May 15 '24 19:05 khluu

I tried having it depends on the block step but that didn't really work.

why? that is how it supposed to work?

aslonnie avatar May 15 '24 23:05 aslonnie

@aslonnie You can take a look at this commit: https://github.com/ray-project/ray/commit/30fdfe8c5fe4252ee304e64d2e6002189310077e which I set mac sanity check step to depend on its block step. The block step is blocked by the step before it though for some reason: https://buildkite.com/ray-project/release-automation/builds/474/dag

When I unblocked the step, the mac sanity check steps didn't even run or wait for agent .. they were just stuck there

khluu avatar May 16 '24 07:05 khluu

When I unblocked the step, the mac sanity check steps didn't even run or wait for agent .. they were just stuck there

you need to add an empty depends_on for the wait/block step too. otherwise, a wait/block step depends on all steps before it.

aslonnie avatar May 16 '24 18:05 aslonnie