Remove ucx-proc dependency and update docs

Since UCX 1.14.1, the `ucx-proc` package is not required anymore and all
UCX builds provide CUDA support. Remove that dependency and update docs
and warnings accordingly.

conda/environments/builddocs.yml

@@ -11,7 +11,6 @@ dependencies:
 - sphinxcontrib-websupport
 - nbsphinx
 - numpydoc
-- ucx-proc=*=cpu
 - recommonmark
 - pandoc=<2.0.0
 - pip

docs/source/install.rst

@@ -35,44 +35,24 @@ support is active by checking for the presence of ``/dev/infiniband/rdma_cm`` an
 Conda
 -----
 
-Some preliminary Conda packages can be installed as so. Replace
-``<CUDA version>`` with the desired version (minimum ``11.0``). These are
-available both on ``rapidsai`` and ``rapidsai-nightly``. Starting with the
-UCX 1.14.1 conda-forge package, InfiniBand support is available again via
-rdma-core, thus building UCX from source is not required solely for that
-purpose anymore but may still be done if desired (e.g., to test for new
-capabilities or bug fixes).
-
-With GPU support (UCX 1.14.0 and higher):
-
-::
-
-    conda create -n ucx -c conda-forge -c rapidsai \
-      cudatoolkit=<CUDA version> ucx>=1.14.0 ucx-py python=3.9
-
-With GPU support (UCX>=1.11.1 and UCX<1.14):
+Conda packages can be installed as so. Replace ``<CUDA version>`` with the
+desired version (minimum ``11.2``). These are available both on ``rapidsai``
+and ``rapidsai-nightly``. Starting with the UCX 1.14.1 conda-forge package,
+InfiniBand support is available again via rdma-core, thus building UCX
+from source is not required solely for that purpose anymore but may still
+be done if desired (e.g., to test for new capabilities or bug fixes).
 
 ::
 
     conda create -n ucx -c conda-forge -c rapidsai \
-      cudatoolkit=<CUDA version> ucx-proc=*=gpu ucx<1.14 ucx-py python=3.9
-
-Without GPU support:
-
-::
-
-    conda create -n ucx -c conda-forge -c rapidsai \
-      ucx-proc=*=cpu ucx ucx-py python=3.9
+      cudatoolkit=<CUDA version> ucx-py
 
 
 Source
 ------
 
 The following instructions assume you'll be using UCX-Py on a CUDA enabled system and is in a `Conda environment <https://docs.conda.io/projects/conda/en/latest/>`_.
 
-.. note::
-    UCX conda-force package older than UCX 1.14.1 do not include InfiniBand support.
-
 
 Build Dependencies
 ~~~~~~~~~~~~~~~~~~
@@ -81,7 +61,11 @@ Build Dependencies
 
     conda create -n ucx -c conda-forge \
         automake make libtool pkg-config \
-        "python=3.9" setuptools "cython>=0.29.14,<3.0.0a0"
+        "python=3.9" setuptools "cython>=3.0.0"
+
+.. note::
+    The Python version must be explicitly specified here, UCX-Py currently supports
+    only Python 3.9 and 3.10.
 
 Test Dependencies
 ~~~~~~~~~~~~~~~~~

docs/source/quickstart.rst

@@ -5,19 +5,12 @@ Quickstart
 Setup
 -----
 
-With GPU support:
+Create a new conda environment with UCX-Py:
 
 ::
 
     conda create -n ucx -c conda-forge -c rapidsai \
-      cudatoolkit=<CUDA version> ucx-proc=*=gpu ucx ucx-py python=3.9
-
-Without GPU support:
-
-::
-
-    conda create -n ucx -c conda-forge -c rapidsai \
-      ucx-proc=*=cpu ucx ucx-py python=3.9
+      cudatoolkit=<CUDA version> ucx-py
 
 For a more detailed guide on installation options please refer to the :doc:`install` page.
 

ucp/_libs/transfer_am.pyx

@@ -117,9 +117,9 @@ def am_send_nbx(
     if buffer.cuda and not ep.worker._context.cuda_support:
         raise ValueError(
             "UCX is not configured with CUDA support, please add "
-            "`cuda_copy` and/or `cuda_ipc` to the UCX_TLS environment"
-            "variable and that the ucx-proc=*=gpu package is "
-            "installed. See "
+            "`cuda_copy` and/or `cuda_ipc` to the `UCX_TLS` environment"
+            "variable if you're manually setting a different value. If you"
+            "are building UCX from source, please see "
             "https://ucx-py.readthedocs.io/en/latest/install.html for "
             "more information."
         )

ucp/_libs/transfer_stream.pyx

@@ -73,9 +73,9 @@ def stream_send_nb(
     if buffer.cuda and not ep.worker._context.cuda_support:
         raise ValueError(
             "UCX is not configured with CUDA support, please add "
-            "`cuda_copy` and/or `cuda_ipc` to the UCX_TLS environment"
-            "variable and that the ucx-proc=*=gpu package is "
-            "installed. See "
+            "`cuda_copy` and/or `cuda_ipc` to the `UCX_TLS` environment"
+            "variable if you're manually setting a different value. If you"
+            "are building UCX from source, please see "
             "https://ucx-py.readthedocs.io/en/latest/install.html for "
             "more information."
         )
@@ -204,9 +204,9 @@ def stream_recv_nb(
     if buffer.cuda and not ep.worker._context.cuda_support:
         raise ValueError(
             "UCX is not configured with CUDA support, please add "
-            "`cuda_copy` and/or `cuda_ipc` to the UCX_TLS environment"
-            "variable and that the ucx-proc=*=gpu package is "
-            "installed. See "
+            "`cuda_copy` and/or `cuda_ipc` to the `UCX_TLS` environment"
+            "variable if you're manually setting a different value. If you"
+            "are building UCX from source, please see "
             "https://ucx-py.readthedocs.io/en/latest/install.html for "
             "more information."
         )

ucp/_libs/transfer_tag.pyx

@@ -70,9 +70,9 @@ def tag_send_nb(
     if buffer.cuda and not ep.worker._context.cuda_support:
         raise ValueError(
             "UCX is not configured with CUDA support, please add "
-            "`cuda_copy` and/or `cuda_ipc` to the UCX_TLS environment"
-            "variable and that the ucx-proc=*=gpu package is "
-            "installed. See "
+            "`cuda_copy` and/or `cuda_ipc` to the `UCX_TLS` environment"
+            "variable if you're manually setting a different value. If you"
+            "are building UCX from source, please see "
             "https://ucx-py.readthedocs.io/en/latest/install.html for "
             "more information."
         )
@@ -217,9 +217,9 @@ def tag_recv_nb(
         if not cuda_support:
             raise ValueError(
                 "UCX is not configured with CUDA support, please add "
-                "`cuda_copy` and/or `cuda_ipc` to the UCX_TLS environment"
-                "variable and that the ucx-proc=*=gpu package is "
-                "installed. See "
+                "`cuda_copy` and/or `cuda_ipc` to the `UCX_TLS` environment"
+                "variable if you're manually setting a different value. If you"
+                "are building UCX from source, please see "
                 "https://ucx-py.readthedocs.io/en/latest/install.html for "
                 "more information."
             )