libwholegraph: build wheels without build isolation#388
libwholegraph: build wheels without build isolation#388rapids-bot[bot] merged 5 commits intorapidsai:mainfrom
Conversation
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
| jobs: | ||
| pr-builder: | ||
| needs: | ||
| - check-nightly-ci |
There was a problem hiding this comment.
Doing a first run of only wheel builds, to test and to populate the cache.
|
/ok to test |
|
/ok to test |
This comment was marked as off-topic.
This comment was marked as off-topic.
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
| sha: ${{ inputs.sha }} | ||
| date: ${{ inputs.date }} | ||
| node_type: cpu8 | ||
| node_type: cpu4 |
There was a problem hiding this comment.
As a note: for some reason cpu4 has tended to queue for a long time in the past. I've been sticking to cpu8 at minimum. Not sure if this is still the case.
There was a problem hiding this comment.
Oh interesting, I didn't realize that. I was just thinking "small = cheap" and hadn't considered the queue.
I just checked and it looks like we allow up to 1000 cpu4 instances at a time (per CPU architecture... 1000 amd64 and 1000 arm64) and haven't come close to hitting that in the last 7 days. The 7-day max looks like around 800 busy runners, and the typical load is usually closer to like 350 busy at a time.
(can share dashboard links privately if you're curious)
So I think this change should be ok and hopefully shouldn't introduce new queuing problems.
|
/merge |
Followup to #182, related to #386 (review)
Proposes using
--no-build-isolationforlibwholegraphwheel builds.--no-build-isolationpreserves all of the source file paths, which should lead to bettersccachehit rate and faster builds.All of the other
lib{something}wheels in RAPIDS are built this way, see rapidsai/build-planning#108Also switches
cugraph-pygbuilds to smaller runners (cpu4). That's a pure-Python package that shouldn't need many resources to build.