<@U06N1KAUCUV> <https://github.com/Eventual-Inc/Da...
# daft-dev
d
@Kevin Wang https://github.com/Eventual-Inc/Daft/pull/4410 might help our poor github runners
k
do we have like a runner quota?
A concern I have about this is that what could end up happening is that we merge in successive PRs and the tests get cancelled, so any errors may show up late
d
no but it feels like we have a limited pool?
A concern I have about this is that what could end up happening is that we merge in successive PRs and the tests get cancelled, so any errors may show up late
I think we can also change this so that it only cancels on new pushes on PRs, instead of all github refs
Copy code
concurrency:
  group: ${{ github.workflow }}-${{ github.event.pull_request.number }}
  cancel-in-progress: true
?
k
d
actually I don't know if the current one would even cancel post merge runs
k
Cancelling the previous tests on a new commit to a PR is a good idea
d
I think with what's in the current PR, successive commits won't cancel each others' runs. Let me double check
(successive commits to main)
Ah my b,
main
is considered a
github.ref
, so yeah it should be removed from the PR. Done
s
FYI earlier today I removed the py-runner tests for this reason
k
If github actions is starting to be a noticeable hinderance to dev velocity we should look into third party CI runners
d
yeah, we should start paying money. It's kinda annoying waiting for actions to kick off sometimes
some other things that would be nice: tags for force merge permissions, better test discoverability (a docs change shouldn't kick off the whole PR test suite)
s
what about using the ARM runners? they're generally faster
k
feel free to play around with them
d
now that you mention it, pre-merge could use ARM, but post merge could test on both ARM and x86
other post merge tests: ASAN and TSAN would be nice
plus one 1
s
Whats with the aqe flag and testing when both enabled and disabled? Do we need to test with both?
d
Hmm that one kinda feels like it would be good to test in both modes
but also feels like it's possible to isolate it via test discovery
👍 1