FIX-#7277: Remove Cudf storage format as unmaintained (#7290)

Signed-off-by: Anatoly Myachev <[email protected]>

Author: anmyachev

docs/conf.py

@@ -27,8 +27,6 @@ def noop_decorator(*args, **kwargs):
 
 # fake modules if they're missing
 for mod_name in (
-    "cudf",
-    "cupy",
     "xgboost",
     "unidist",
     "unidist.config",
@@ -39,10 +37,6 @@ def noop_decorator(*args, **kwargs):
         sys.modules[mod_name] = types.ModuleType(
             mod_name, f"fake {mod_name} for building docs"
         )
-if not hasattr(sys.modules["cudf"], "DataFrame"):
-    sys.modules["cudf"].DataFrame = type("DataFrame", (object,), {})
-if not hasattr(sys.modules["cupy"], "ndarray"):
-    sys.modules["cupy"].ndarray = type("ndarray", (object,), {})
 if not hasattr(sys.modules["xgboost"], "Booster"):
     sys.modules["xgboost"].Booster = type("Booster", (object,), {})
 if not hasattr(sys.modules["unidist"], "remote"):

docs/development/architecture.rst

@@ -223,10 +223,6 @@ documentation page on :doc:`contributing </development/contributing>`.
     - Uses native python execution - mainly used for debugging.
     - The storage format is `pandas` and the in-memory partition type is a pandas DataFrame.
     - For more information on the execution path, see the :doc:`pandas on Python </flow/modin/core/execution/python/implementations/pandas_on_python/index>` page.
-- cuDF on Ray (experimental)
-    - Uses the Ray_ execution framework.
-    - The storage format is `cudf` and the in-memory partition type is a cuDF DataFrame.
-    - For more information on the execution path, see the :doc:`cuDF on Ray </flow/modin/core/execution/ray/implementations/cudf_on_ray/index>` page.
 
 .. _directory-tree:
 
@@ -300,7 +296,6 @@ details. The documentation covers most modules, with more docs being added every
    │   │   │   │   ├───common
    │   │   │   │   ├─── :doc:`generic </flow/modin/core/execution/ray/generic>`
    │   │   │   │   └───implementations
-   │   │   │   │       ├─── :doc:`cudf_on_ray </flow/modin/core/execution/ray/implementations/cudf_on_ray/index>`
    │   │   │   │       └─── :doc:`pandas_on_ray </flow/modin/core/execution/ray/implementations/pandas_on_ray/index>`
    │   │   │   └───unidist
    │   │   │       ├───common
@@ -310,7 +305,6 @@ details. The documentation covers most modules, with more docs being added every
    │   │   ├─── :doc:`io </flow/modin/core/io/index>`
    │   │   └─── :doc:`storage_formats </flow/modin/core/storage_formats/index>`
    │   │       ├─── :doc:`base </flow/modin/core/storage_formats/base/query_compiler>`
-   │   │       ├───cudf
    │   │       └─── :doc:`pandas </flow/modin/core/storage_formats/pandas/index>`
    │   ├───distributed
    │   │   ├───dataframe

docs/flow/modin/core/dataframe/index.rst

@@ -29,7 +29,6 @@ Engine specific:
 Execution system specific:
 
 * :doc:`Modin PandasOnRayDataframe </flow/modin/core/execution/ray/implementations/pandas_on_ray/index>` is a specialization of the Core Modin Dataframe for ``PandasOnRay`` execution.
-* :doc:`Modin cuDFOnRayDataframe </flow/modin/core/execution/ray/implementations/cudf_on_ray/index>` is a specialization of the Core Modin Dataframe for ``cuDFOnRay`` execution.
 * :doc:`Modin PandasOnDaskDataframe </flow/modin/core/execution/dask/implementations/pandas_on_dask/index>` is specialization of the Core Modin Dataframe for ``PandasOnDask`` execution.
 * :doc:`Modin PandasOnPythonDataframe </flow/modin/core/execution/python/implementations/pandas_on_python/index>` is a specialization of the Core Modin Dataframe for ``PandasOnPython`` execution.
 * :doc:`Modin PandasOnUnidistDataframe </flow/modin/core/execution/unidist/implementations/pandas_on_unidist/index>` is a specialization of the Core Modin Dataframe for ``PandasOnUnidist`` execution.

docs/flow/modin/core/execution/ray/implementations/cudf_on_ray/dataframe.rst

@@ -64,9 +64,5 @@ High-level module overview
 
 This module houses submodules of all of the stable storage formats:
 
-..
-    TODO: Insert a link to <cuDF module> when it is added (issue #3323)
-
 - :doc:`Base module <base/query_compiler>` contains an abstract query compiler class which defines common API.
 - :doc:`Pandas module <pandas/index>` contains query compiler and text parsers for pandas storage format.
-- cuDF module contains query compiler and text parsers for cuDF storage format.

docs/flow/modin/core/execution/ray/implementations/cudf_on_ray/index.rst

@@ -257,7 +257,7 @@ class StorageFormat(EnvironmentVariable, type=str):
 
     varname = "MODIN_STORAGE_FORMAT"
     default = "Pandas"
-    choices = ("Pandas", "Cudf")
+    choices = ("Pandas",)
 
 
 class IsExperimental(EnvironmentVariable, type=bool):
@@ -412,10 +412,7 @@ def _get_default(cls) -> int:
         -------
         int
         """
-        if StorageFormat.get() == "Cudf":
-            return GpuCount.get()
-        else:
-            return CpuCount.get()
+        return CpuCount.get()
 
     @classmethod
     def get(cls) -> int:

docs/flow/modin/core/execution/ray/implementations/cudf_on_ray/io.rst

@@ -26,7 +26,6 @@
 import pandas
 from pandas.util._decorators import doc
 
-from modin.config import IsExperimental
 from modin.core.io import BaseIO
 from modin.utils import get_current_execution
 
@@ -778,19 +777,3 @@ def prepare(cls):
         )
 
         cls.io_cls = PandasOnUnidistIO
-
-
-# EXPERIMENTAL FACTORIES
-# Factories that operate only in experimental mode. They provide access to executions
-# that have little coverage of implemented functionality or are not stable enough.
-@doc(_doc_factory_class, execution_name="cuDFOnRay")
-class ExperimentalCudfOnRayFactory(BaseFactory):
-    @classmethod
-    @doc(_doc_factory_prepare_method, io_module_name="``cuDFOnRayIO``")
-    def prepare(cls):
-        from modin.core.execution.ray.implementations.cudf_on_ray.io import cuDFOnRayIO
-
-        if not IsExperimental.get():
-            raise ValueError("'CudfOnRay' only works in experimental mode.")
-
-        cls.io_cls = cuDFOnRayIO

docs/flow/modin/core/execution/ray/implementations/cudf_on_ray/partitioning/axis_partition.rst

@@ -34,7 +34,6 @@
     RayInitCustomResources,
     RayRedisAddress,
     RayRedisPassword,
-    StorageFormat,
     ValueSource,
 )
 from modin.core.execution.utils import set_env
@@ -139,17 +138,6 @@ def initialize_ray(
             with set_env(**env_vars):
                 ray.init(**ray_init_kwargs)
 
-        if StorageFormat.get() == "Cudf":
-            from modin.core.execution.ray.implementations.cudf_on_ray.partitioning import (
-                GPU_MANAGERS,
-                GPUManager,
-            )
-
-            # Check that GPU_MANAGERS is empty because _update_engine can be called multiple times
-            if not GPU_MANAGERS:
-                for i in range(GpuCount.get()):
-                    GPU_MANAGERS.append(GPUManager.remote(i))
-
     # Now ray is initialized, check runtime env config - especially useful if we join
     # an externally pre-configured cluster
     runtime_env_vars = ray.get_runtime_context().runtime_env.get("env_vars", {})
@@ -162,11 +150,7 @@ def initialize_ray(
                 )
 
     num_cpus = int(ray.cluster_resources()["CPU"])
-    num_gpus = int(ray.cluster_resources().get("GPU", 0))
-    if StorageFormat.get() == "Cudf":
-        NPartitions._put(num_gpus)
-    else:
-        NPartitions._put(num_cpus)
+    NPartitions._put(num_cpus)
 
     # TODO(https://github.com/ray-project/ray/issues/28216): remove this
     # workaround once Ray gives a better way to suppress task errors.