Skip to content

CI: invalid dependencies in nightly wheel-tests-cudf / oldest-deps jobs #25224

Closed

Description

@TomAugspurger

https://github.com/NVIDIA/cudf/actions/runs/43163545088/job/131000110370#step:13:747

ImportError while loading conftest '/__w/cudf/cudf/python/cudf/cudf/tests/conftest.py'.
conftest.py:21: in <module>
    import cudf
/pyenv/versions/3.12.16/lib/python3.12/site-packages/cudf/__init__.py:18: in <module>
    from cudf import api, core, datasets, testing
/pyenv/versions/3.12.16/lib/python3.12/site-packages/cudf/api/__init__.py:4: in <module>
    from cudf.api import extensions, types
/pyenv/versions/3.12.16/lib/python3.12/site-packages/cudf/api/extensions/__init__.py:6: in <module>
    from cudf.api.extensions.accessor import (
/pyenv/versions/3.12.16/lib/python3.12/site-packages/cudf/api/extensions/accessor.py:6: in <module>
    from cudf.core.dataframe import DataFrame
/pyenv/versions/3.12.16/lib/python3.12/site-packages/cudf/core/dataframe.py:33: in <module>
    from numba_cuda_mlir import cuda
/pyenv/versions/3.12.16/lib/python3.12/site-packages/numba_cuda_mlir/__init__.py:43: in <module>
    from numba_cuda_mlir.numba_cuda.np.numpy_support import carray, farray  # noqa: F401
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
/pyenv/versions/3.12.16/lib/python3.12/site-packages/numba_cuda_mlir/numba_cuda/__init__.py:32: in <module>
    from .device_init import *
/pyenv/versions/3.12.16/lib/python3.12/site-packages/numba_cuda_mlir/numba_cuda/device_init.py:6: in <module>
    from numba_cuda_mlir.numba_cuda import cg
/pyenv/versions/3.12.16/lib/python3.12/site-packages/numba_cuda_mlir/numba_cuda/cg.py:5: in <module>
    from numba_cuda_mlir.numba_cuda.extending import overload, overload_method
/pyenv/versions/3.12.16/lib/python3.12/site-packages/numba_cuda_mlir/numba_cuda/extending.py:17: in <module>
    from numba_cuda_mlir.numba_cuda.typing.typeof import typeof_impl  # noqa: F401
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
/pyenv/versions/3.12.16/lib/python3.12/site-packages/numba_cuda_mlir/numba_cuda/typing/__init__.py:10: in <module>
    from .context import BaseContext, Context
/pyenv/versions/3.12.16/lib/python3.12/site-packages/numba_cuda_mlir/numba_cuda/typing/context.py:15: in <module>
    from numba_cuda_mlir.numba_cuda.typing.typeof import typeof, Purpose
/pyenv/versions/3.12.16/lib/python3.12/site-packages/numba_cuda_mlir/numba_cuda/typing/typeof.py:20: in <module>
    from cuda.core import Buffer
/pyenv/versions/3.12.16/lib/python3.12/site-packages/cuda/core/__init__.py:93: in <module>
    _import_versioned_module()
/pyenv/versions/3.12.16/lib/python3.12/site-packages/cuda/core/__init__.py:74: in _import_versioned_module
    _bindings_floor.check_installed_bindings(
/pyenv/versions/3.12.16/lib/python3.12/site-packages/cuda/core/_bindings_floor.py:204: in check_installed_bindings
    raise ImportError(
E   ImportError: cuda.core 1.3.0 requires cuda-bindings >= 12.9.8 for CUDA 12, but cuda-bindings 12.9.3 is installed. Upgrade with: pip install -U "cuda-bindings>=12.9.8,<13"
Error: Process completed with exit code 4.

We need to figure out what we care most about. If want cuda-bindings==12.9.3 then we'll need to pin cuda-core too.

Also wheel-tests-cudf / 12.2.2, 3.12, arm64, linux, rockylinux8, a100, latest-driver, oldest-deps, unit-tests-cudf-pandas / 12.9.2, 3.12, amd64, linux, rockylinux8, l4, latest-driver, oldest-deps, and unit-tests-cudf-pandas / 12.2.2, 3.12, arm64, linux, rockylinux8, a100, latest-driver, oldest-deps.

Finally we have issues in the upstream polars test env, https://github.com/NVIDIA/cudf/actions/runs/43163545088/job/131000089134. IIUC, torch downgraded our cuda-bindings, but not cuda-core.

Activity

  1. brandon-b-miller commented on Nov 12, 2026

    @brandon-b-miller
    Contributor

    What's the context around requiring the cuda-bindings oldest dep to float back to 12.9.3? Can the oldest cuda-bindings dep be 12.9.8?

  2. TomAugspurger commented on Nov 12, 2026

    @TomAugspurger
    ContributorAuthor

    Looks like #21928 / NVIDIA/cuda-python#2743. I suspect we'd be OK with bumping that to 12.9.8. I can work on that (targeting release/26.12 I assume).

  3. bdice commented on Nov 12, 2026

    @bdice
    Contributor

    Yes, it's fine to bump the minimum requirement to 12.9.8. Let's do that for release/26.12.

  4. TomAugspurger commented on Nov 12, 2026

    @TomAugspurger
    ContributorAuthor

    Also issues in the polars envs: https://github.com/NVIDIA/cudf/actions/runs/43163545088/job/131000089134

    Likely from torch downgrading cuda-bindings, but not cuda-core.

  5. added a commit that references this issue on Nov 12, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    cudf.pandasIssues specific to cudf.pandas

    Type

    No type

    Projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions