Update Github Actions#1385
Conversation
|
Running full CI to make sure there's no problems with the updates. |
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## develop #1385 +/- ##
========================================
Coverage 90.68% 90.68%
========================================
Files 135 135
Lines 14663 14663
========================================
Hits 13297 13297
Misses 1366 1366 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
|
The changes look fine, do you know if the two failed jobs were just a fluke or? |
I don't think the failures are a fluke. The same jobs failed 3 times due to timing out (all were cancelled at ~6hrs). Looking into it... |
|
Oh okay, yeah its probably that conda line change. This is currently blocking all CI runs. |
Yeah, probably related to the conda/setup-miniconda change, but the log shows pytest does start: it collects 6102 tests and creates the xdist workers before the job times out. So I’m not sure yet whether we’re hanging on a particular test or just running much slower under the new macOS conda env. I’m switching the pytest line to verbose so we can see where it stalls. |
|
Something is definitely wrong with it. Its taking hours to get through tests that are usually instant.... |
Yeah, I see that. One of the main changes that I'm seeing from |
|
Setting |
|
Ahh sorry, I just got your prior github message. I see. The conda platform changed from intel to arm. I see. I think this is the same issue that I hit recently for openblas... just with the other packages now.... .... |
|
Can you check if that env is trying to install (ie emulate ) x86_64 libraries on apple silicon? This is what was the problem last time... |
Thanks. I'll check... |
I'm not seeing any x86_64 installs in the environment log. Everything is arm64 or universal2. |
|
looking at the runner specs, the change from intel to arm goes from 4 cores to 3. Maybe that could be the problem. |
|
Weren't we already using this runner before this code change. Like if you put in the "force use the old node" line in the |
We were using the same runner, but the new version updated the conda architecture to arm64.
There was a single commit for the change to the conda line. Looking at the environment logs before and after you can see the platform change. Also, the tests passed and ran in expected time prior to changing that single line. |
|
Cool, so nothing changed about the runner. I don't doubt the conda platform change, but that is software, not a runner change, core count change, etc. That was all fine, and I think can be confirmed to still be fine as I suggested. A concern here is that this indicates there might be a problem installing and running the code in a relatively normal way in a conda env due to some yet to be identified software issues. Just disabling OMP threading might be a workaround, but isn't identifying or solving the issue in any meaningful way considering it was working reasonably well before .... and isn't what I would recommend to users either outside special circumstances.... Considering arm is virtually all the macs now I'm not inclined to accept ignoring it. |
garrettwrong
left a comment
garrettwrong
left a comment
There was a problem hiding this comment.
node 24 is going to be so worth it
|
Looks good, thanks! |
|
Did you want to merge this or there is still more? |
Merging |
2 participants
Update Github actions to support node.js 24