-
Notifications
You must be signed in to change notification settings - Fork 892
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[BUG] Builds are not reproducible #11533
Comments
If you need to test with a specific version of rapids-cmake you can follow this guide https:/rapidsai/rapids-cmake#overriding-rapidscmake which shows how specify a specific version ( GIT_TAG can also be a SHA ). You would place this in the |
@robertmaynard I see that this was closed and then re-opened again. Are there plans to do more for this? The docs you pointed us to I think are workable, and we can probably patch CUDF as we build it to get a consistent version. But if you are going to do more, like expose the version through a CMAKE variable or something it would be good to understand that, because I don't want to patch CUDF unless I have to. |
I accidently closed, and didn't want to come across as presumptious so I re-opened. The balance between ease of use and reproducibility for 'nightly' builds is hard, and we currently go towards ease of use. We could move over to explicit SHA's but we would need to develop some infrastructure first. Primarily we need a robust way to ensure that during code-freeze all projects have synced to the same rapids-cmake SHA. |
I was thinking of something very simple. Like an environment variable for the SHA and if it is not set, then it defaults to what happens today. I realize that for most of rapids and CUDF in particular that ease of use is paramount. All I would like is a way to be able to set this without having to patch CUDF itself. I could even write up the patch myself. I mostly want to have someones blessing that this is acceptable. |
The ability to pin rapids-cmake is coming to all consumers via rapidsai/rapids-cmake#257 which will allow you to pin by passing |
This issue has been labeled |
Closing since rapidsai/rapids-cmake#257 has been merged |
@robertmaynard We reopened this issue following discussion with the Spark team. Is there work to be done for cudf specifically, or would all the work be in |
No work in libcudf, all the work will be in |
I'm closing this since I believe all the requirements were addressed in rapidsai/rapids-cmake#530 and its follow-ups. @robertmaynard please reopen if I am mistaken. |
You are correct. |
Describe the bug
I tried to redo an older 22.10 build from a specific SHA hash and it failed. see NVIDIA/spark-rapids-jni#479 It looks like cudf always pulls the latest version of rapids-cmake. So if rapids-cmake ever changes older builds are impacted and we have no way to keep the exact version of rapids-cmake.
Expected behavior
I check out a specific version of cudf from git it always builds exactly the same. Code does not magically change on me without my knowledge.
If CUDF always wants to run on the latest version, that is fine. But then we need a way to be able to set the exact SHA version we want for rapids-cmake.
The text was updated successfully, but these errors were encountered: