WIP: [DO NOT MERGE] test gha-tools changes

[jameslamb]

Description

Using this together with rapidsai/rmm#1909 to test documentation updates describing how to reproduce CI locally now that RAPIDS projects are using the GitHub Actions artifact store for conda packages and wheels (e.g. #18539).

Testing these, specifically:

• document how to use CI artifacts with GitHub Actions artifact store docs#601
• WIP: [NOT READY FOR REVIEW] reduce duplication in rapids-get-pr-* scripts gha-tools#173

Using inputs from:

• WIP: [DO NOT MERGE] test changes to GitHub Actions configs and scripts rmm#1909

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[jameslamb]

Ok so this did "work" from the perspective of downloading rmm and librmm packages from the expected run (https://github.com/rapidsai/rmm/actions/runs/14937027027, on rapidsai/rmm#1909).

But those packages were ignored in builds, in favor of nightlies from rapidsai-nightly.

In a conda build, for example:

Downloading and decompressing rmm_conda_cpp_cuda12_x86_64 from Run ID 14937027027 into /tmp/tmp.s9B1FY59sJ
Downloading and decompressing rmm_conda_python_cuda12_py310_x86_64 from Run ID 14937027027 into /tmp/tmp.wqsVtjko7i
...
 │ │ │ librmm                           ┆                                          ┆ 25.06.00a32  ┆ cuda12_250509_dbd8cc7a       ┆ rapidsai-nightly ┆   1.23 MiB │

(build link)

I think that's showing us that adding strict channel priority (rapidsai/build-planning#84) means the instructions for using CI artifacts across PRs (these, in rapidsai/docs) no longer work.

If we want these downloaded artifacts to be used in favor of nightlies, the local channels need to be prepended to the list of channels considered by rattler-build, not merely added somewhere in the mix with conda config --system --add.

[jameslamb]

/ok to test a0f2cfd

[jameslamb]

/ok to test

[jameslamb]

This is mostly working ... definitely well enough to move forward with rapidsai/docs#601 and rapidsai/gha-tools#173, in my opinion.

With the proposed patterns here + in the linked PRs above, it should be possible to get 100% confidence that the PR CI artifacts will be used for:

• conda builds
• wheel builds
• wheel tests

Conda tests can sometimes fall back to nightlies, because we're not yet using strict channel priority in conda test jobs. Doing that in the future would be helpful: rapidsai/build-planning#14 (comment)

Looking to see rmm artifacts from run 15006618039 (https://github.com/rapidsai/rmm/actions/runs/15006618039/job/42166726968?pr=1909).

✅ `conda-cpp-build` (using CI artifacts)

Downloading and decompressing rmm_conda_cpp_cuda12_x86_64 from Run ID 15006618039 into /tmp/tmp.Is9tAcqNRd
Downloading and decompressing rmm_conda_python_cuda12_py[31](https://github.com/rapidsai/cudf/actions/runs/15009358240/job/42175206557?pr=18747#step:9:32)0_x86_64 from Run ID 15006618039 into /tmp/tmp.n8mUJGq0pe
...
 │ │ │ librmm               ┆ 25.06.00a44  ┆ cuda12_250513_b546c0c7 ┆ tmp.Is9tAcqNRd   ┆   1.24 MiB │
...

(build link)

✅ `conda-python-build` (using CI artifacts)

conda-python-build

Downloading and decompressing rmm_conda_cpp_cuda12_x86_64 from Run ID 15006618039 into /tmp/tmp.I63b0pgzRn
Downloading and decompressing rmm_conda_python_cuda12_py313_x86_64 from Run ID 15006618039 into /tmp/tmp.Zn5Y0176ET
...
 │ │ │ librmm                           ┆ 25.06.00a44  ┆ cuda12_250513_b546c0c7       ┆ tmp.I63b0pgzRn   ┆   1.24 MiB │
...

(build link)

⚠️ `conda-cpp-tests` (fell back to older nightlies)

The rmm CI artifacts are getting downloaded:

Downloading and decompressing rmm_conda_cpp_cuda12_x86_64 from Run ID 15006618039 into /tmp/tmp.5r8ZTtsbKl
Downloading and decompressing rmm_conda_python_cuda12_py313_x86_64 from Run ID 15006618039 into /tmp/tmp.nJnU812v8u

And the channels added to the conda env:

channels:
- /tmp/tmp.xU2VE7qMqe
- rapidsai
- rapidsai-nightly
- dask/label/dev
- conda-forge
- nvidia
dependencies:
...

But conda is still choosing nightlies instead:

...
  + librmm                 25.06.00a36  cuda12_250513_b30828d0  rapidsai-nightly          1MB
...

(build link)

✅ `conda-python-cudf-tests` (using CI artifacts)

Downloading and decompressing rmm_conda_cpp_cuda12_x86_64 from Run ID 15006618039 into /tmp/tmp.lvggUAdouZ
Downloading and decompressing rmm_conda_python_cuda12_py313_x86_64 from Run ID 15006618039 into /tmp/tmp.DUaZqWXGda
...
channels:
- /tmp/tmp.g817QzqT2W
- /tmp/tmp.Qb848l23FM
- rapidsai
- rapidsai-nightly
- dask/label/dev
- conda-forge
- nvidia
dependencies:
...
  + librmm                                25.06.00a44  cuda12_250513_b546c0c7          /tmp/tmp.lvggUAdouZ       1MB
...
librmm                    25.06.00a44     cuda12_250513_b546c0c7    file:///tmp/tmp.lvggUAdouZ

(build link)

✅ `wheel-build-libcudf` (using CI artifacts)

Downloading and decompressing rmm_wheel_cpp_librmm_cu12_x86_64 from Run ID 15006618039 into /tmp/tmp.m2QBJjxJnZ
Downloading and decompressing rmm_wheel_python_rmm_cu12_py313_x86_64 from Run ID 15006618039 into /tmp/tmp.1QwgTU9GV9
...
Processing /tmp/tmp.m2QBJjxJnZ/librmm_cu12-25.6.0a44-py3-none-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (from -r /tmp/requirements-build.txt (line 7))

(build link)

✅ `wheel-build-pylibcudf` (using CI artifacts)

Downloading and decompressing rmm_wheel_cpp_librmm_cu12_x86_64 from Run ID 15006618039 into /tmp/tmp.oE968qwu1O
Downloading and decompressing rmm_wheel_python_rmm_cu12_py313_x86_64 from Run ID 15006618039 into /tmp/tmp.DpSpSPQi5p
...
Processing /tmp/tmp.oE968qwu1O/librmm_cu12-25.6.0a44-py3-none-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl
...

(build link)

✅ `wheel-build-cudf` (using CI artifacts)

Downloading and decompressing rmm_wheel_cpp_librmm_cu12_x86_64 from Run ID 15006618039 into /tmp/tmp.qH7VMpGUME
Downloading and decompressing rmm_wheel_python_rmm_cu12_py313_x86_64 from Run ID 15006618039 into /tmp/tmp.OWSwGFJuHA
...
Processing /tmp/tmp.qH7VMpGUME/librmm_cu12-25.6.0a44-py3-none-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl
...

(build link)

✅ `wheel-tests-cudf` (using CI artifacts)

Downloading and decompressing rmm_wheel_cpp_librmm_cu12_x86_64 from Run ID 15006618039 into /tmp/tmp.j4JY3ho3Nw
Downloading and decompressing rmm_wheel_python_rmm_cu12_py310_x86_64 from Run ID 15006618039 into /tmp/tmp.kjoVmTjSIG
...
Processing /tmp/tmp.j4JY3ho3Nw/librmm_cu12-25.6.0a44-py3-none-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (from libcudf-cu12==25.6.0a345)

(build link)

[jameslamb]

Shifting this testing over to branch-25.08, now that rmm is about to enter code freeze. But everything being tested here is on both branch-25.06 and branch-25.08, so this should still hopefully give us decent confidence that things are working for branch-25.06.

[jameslamb]

/ok to test