Enable AWSSDK on Linux by statically linking OpenSSL and cURL

.github/workflows/_build_test_upload.yml

@@ -54,7 +54,7 @@ jobs:
   wheel_build_test:
     needs: get_release_type
     runs-on: ${{ matrix.os }}
-    # container: ${{ startsWith( matrix.os, 'ubuntu' ) && 'quay.io/pypa/manylinux2014_x86_64' || null }}
+    container: ${{ startsWith( matrix.os, 'ubuntu' ) && 'pytorch/manylinux-cpu' || null }}
     strategy:
       fail-fast: false
       matrix:
@@ -69,7 +69,7 @@ jobs:
           - "3.10"
     steps:
       - name: Setup Python ${{ matrix.python-version }}
-        # if: ${{ ! startsWith( matrix.os, 'ubuntu' ) }}
+        if: ${{ ! startsWith( matrix.os, 'ubuntu' ) }}
         uses: actions/setup-python@v2
         with:
           python-version: ${{ matrix.python-version }}
@@ -88,61 +88,72 @@ jobs:
           arch: x64
       - name: Install Build Dependency
         shell: bash
+        env:
+          PYTHON_VERSION: ${{ matrix.python-version }}
         run: |
-          # if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
-          #   yum -y install ninja-build
-          #   yum -y install openssl-devel openssl-static curl-devel zlib-devel
-          # else
-          pip install cmake ninja
-          echo "/home/runner/.local/bin" >> $GITHUB_PATH
-          # fi
+          if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
+            source packaging/manylinux/python_helper.sh
+            yum -y install ninja-build zlib-static
+            # Docker path is /__w by default
+            export WORKSPACE="/__w"
+            # Install static OpenSSL/libcrypto library
+            ./packaging/manylinux/install_openssl_curl.sh
+          else
+            pip install cmake ninja
+            echo "/home/runner/.local/bin" >> $GITHUB_PATH
+          fi
       - name: Install PyTorch and Build TorchData Wheel
         shell: bash
         env:
           PYTHON_VERSION: ${{ matrix.python-version }}
           PYTORCH_VERSION: ${{ inputs.pytorch_version }}
-          BUILD_S3: ${{ startsWith( matrix.os, 'ubuntu' ) && 'OFF' || 'ON' }}
+          BUILD_S3: 1
         run: |
-          # if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
-          #   # See: https://github.com/actions/checkout/issues/760
-          #   git config --global --add safe.directory /__w/data/data
-          #   source packaging/manylinux_wheel_helper.sh
-          # fi
+          if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
+            source packaging/manylinux/python_helper.sh
+            # See: https://github.com/actions/checkout/issues/760
+            git config --global --add safe.directory "$WORKSPACE/data/data"
+            # AWSSDK uses $CMAKE_PREFIX_PATH to find openssl
+            export OPENSSL_ROOT_DIR="/__w/ssl"
+            export CURL_ROOT_DIR="/__w/curl"
+            export CMAKE_PREFIX_PATH="$OPENSSL_ROOT_DIR:$CURL_ROOT_DIR:$CMAKE_PREFIX_PATH"
+            export STATIC_DEPS=TRUE
+          fi
           packaging/build_wheel.sh
       - name: Validate TorchData Wheel
         shell: bash
         env:
           PYTHON_VERSION: ${{ matrix.python-version }}
         run: |
-          # if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
-          #   source packaging/manylinux_wheel_helper.sh
-          #   pip3 install auditwheel
-          # fi
+          if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
+            source packaging/manylinux/python_helper.sh
+            pip3 install auditwheel
+          fi
           pip3 install pkginfo
           for pkg in dist/torchdata*.whl; do
               echo "PkgInfo of $pkg:"
               pkginfo $pkg
-              # if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
-              #   auditwheel show $pkg
-              # fi
+              if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
+                auditwheel show $pkg
+              fi
           done
       - name: Install TorchData Wheel
         shell: bash
         env:
           PYTHON_VERSION: ${{ matrix.python-version }}
         run: |
-          # if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
-          #   source packaging/manylinux_wheel_helper.sh
-          # fi
+          if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
+            source packaging/manylinux/python_helper.sh
+          fi
           pip3 install dist/torchdata*.whl
       - name: Run DataPipes Tests with pytest
         shell: bash
         env:
           PYTHON_VERSION: ${{ matrix.python-version }}
         run: |
-          # if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
-          #   source packaging/manylinux_wheel_helper.sh
-          # fi
+          if ${{ startsWith( matrix.os, 'ubuntu' ) }}; then
+            source packaging/manylinux/python_helper.sh
+          fi
           pip3 install -r test/requirements.txt
           pytest --no-header -v test --ignore=test/test_period.py --ignore=test/test_text_examples.py --ignore=test/test_audio_examples.py
       - name: Upload Wheels to Github

.github/workflows/ci.yml

@@ -29,9 +29,6 @@ jobs:
           - 3.7
           - 3.8
           - 3.9
-        with-s3:
-          - 1
-          - 0
     steps:
       - name: Setup additional system libraries
         if: startsWith( matrix.os, 'ubuntu' )
@@ -44,10 +41,10 @@ jobs:
         with:
           python-version: ${{ matrix.python-version }}
       - name: Setup msbuild on Windows
-        if: matrix.with-s3 == 1 && matrix.os == 'windows-latest'
+        if: matrix.os == 'windows-latest'
         uses: microsoft/[email protected]
       - name: Set up Visual Studio shell
-        if: matrix.with-s3 == 1 && matrix.os == 'windows-latest'
+        if: matrix.os == 'windows-latest'
         uses: egor-tensin/vs-shell@v2
         with:
           arch: x64
@@ -65,7 +62,7 @@ jobs:
         run: |
           python setup.py install
         env:
-          BUILD_S3: ${{ matrix.with-s3 }}
+          BUILD_S3: 1
       - name: Install test requirements
         run: pip3 install -r test/requirements.txt
       - name: Run DataPipes tests with pytest

packaging/README.md

@@ -29,3 +29,16 @@ PYTHON_VERSION=3.9 PYTORCH_VERSION=1.11.0 packaging/build_wheel.sh
 ```bash
 PYTHON_VERSION=3.9 packaging/build_wheel.sh
 ```
+## [`AWSSDK`](https://github.com/aws/aws-sdk-cpp)
+
+The following table is the corresponding `torchdata` binaries with pre-compiled `AWSSDK` extension on different operating systems.
+
+| `torchdata`        | `Wheel`            | `Conda`            |
+| ------------------ | ------------------ | ------------------ |
+| Linux              | :heavy_check_mark: | :heavy_check_mark: |
+| Windows            | :heavy_check_mark: | :x:                |
+| MacOS              | :heavy_check_mark: | :heavy_check_mark: |
+
+### Manylinux
+
+`AWSSDK` requires OpenSSL and cURL. In order to provide `manylinux2014_x86_64` wheels with `AWSSDK` enabled, `torchdata` distributions are bundled with OpenSSL(1.1.1o) and cURL(7.38.1). If anything is out of date, please open an issue to request upgrading them.

packaging/manylinux/install_openssl_curl.sh

@@ -0,0 +1,58 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+OPENSSL_URL="https://www.openssl.org/source/"
+OPENSSL_NAME="openssl-1.1.1o"
+OPENSSL_SHA256="9384a2b0570dd80358841464677115df785edb941c71211f75076d72fe6b438f"
+OPENSSL_BUILD_FLAGS="no-ssl2 no-zlib no-shared no-comp no-dynamic-engine enable-ec_nistp_64_gcc_128"
+
+CURL_URL="https://github.com/curl/curl/releases/download"
+CURL_NAME="curl-7.83.1"
+CURL_BUILD_FLAGS="--disable-shared"
+
+function check_sha256sum {
+    local fname=$1
+    local sha256=$2
+    echo "${sha256}  ${fname}" > ${fname}.sha256
+    sha256sum -c ${fname}.sha256
+    rm ${fname}.sha256
+}
+
+yum erase -y openssl-devel curl-devel
+
+pushd ${WORKSPACE}
+
+# OpenSSL
+curl -fsSL -o ${OPENSSL_NAME}.tar.gz ${OPENSSL_URL}/${OPENSSL_NAME}.tar.gz
+check_sha256sum ${OPENSSL_NAME}.tar.gz ${OPENSSL_SHA256}
+tar zxf ${OPENSSL_NAME}.tar.gz
+
+pushd ${OPENSSL_NAME}
+
+./config $OPENSSL_BUILD_FLAGS --prefix=${WORKSPACE}/ssl --openssldir=${WORKSPACE}/ssl
+make -j4 > /dev/null
+# avoid installing the docs
+# https://github.com/openssl/openssl/issues/6685#issuecomment-403838728
+make install_sw > /dev/null
+
+popd
+rm -rf ${OPENSSL_NAME} ${OPENSSL_NAME}.tar.gz
+
+# cURL
+curl -fsSL -o ${CURL_NAME}.tar.gz ${CURL_URL}/${CURL_NAME//./_}/${CURL_NAME}.tar.gz
+tar zxf ${CURL_NAME}.tar.gz
+
+pushd ${CURL_NAME}
+
+./configure ${CURL_BUILD_FLAGS} --with-openssl=${WORKSPACE}/ssl --prefix=${WORKSPACE}/curl 
+make -j4 > /dev/null
+make install > /dev/null
+
+popd
+rm -rf ${CURL_NAME} ${CURL_NAME}.tar.gz
+
+popd

packaging/torchdata/meta.yaml

@@ -47,9 +47,8 @@ test:
     # The following packages are not on the default conda channel
     # - iopath
     # - rarfile
-  # TODO: Re-enable it after #386 is landed
-  # commands:
-  #   - pytest --no-header -v test --ignore=test/test_period.py --ignore=test/test_text_examples.py --ignore=test/test_audio_examples.py
+  commands:
+    - pytest --no-header -v test --ignore=test/test_period.py --ignore=test/test_text_examples.py --ignore=test/test_audio_examples.py
 
 about:
   home: https://github.com/pytorch/data

third_party/CMakeLists.txt

@@ -56,13 +56,8 @@ else()
     endif()
   endforeach()
 
-  if(WIN32)
-    list(APPEND AWSSDK_LIBRARIES "Userenv;version;ws2_32;Bcrypt;Wininet;winhttp;Crypt32;Secur32;NCrypt;Shlwapi")
-  elseif(APPLE)
-    list(APPEND AWSSDK_LIBRARIES "pthread;curl")
-  else()
+  if(UNIX AND NOT APPLE)
     list(APPEND AWSSDK_LIBRARIES "${aws_cpp_sdk_INSTALL}/lib/libs2n.a")
-    list(APPEND AWSSDK_LIBRARIES "pthread;crypto;ssl;z;curl")
   endif()
 
   include(ExternalProject)

torchdata/csrc/CMakeLists.txt

@@ -11,13 +11,32 @@ if(BUILD_S3)
   # see https://github.com/actions/setup-python/issues/121#issuecomment-1014500503
   set(Python_FIND_FRAMEWORK "LAST")
 
-  if (WIN32)
+  if(WIN32)
       find_package(Python3 ${PYTHON_VERSION} EXACT COMPONENTS Interpreter Development)
     set(ADDITIONAL_ITEMS Python3::Python)
   else()
     find_package(Python3 COMPONENTS Interpreter Development)
   endif()
 
+  # AWSSDK Dependencies
+  if(WIN32)
+    set(AWSSDK_DEP_LIBRARIES "Userenv;version;ws2_32;Bcrypt;Wininet;winhttp;Crypt32;Secur32;NCrypt;Shlwapi")
+  elseif(APPLE)
+    set(AWSSDK_DEP_LIBRARIES pthread curl)
+  elseif(UNIX)
+    set(AWSSDK_DEP_LIBRARIES pthread)
+    if(STATIC_DEPS)
+      set(OPENSSL_USE_STATIC_LIBS TRUE)
+    endif()
+    include(FindZLIB)
+    list(APPEND AWSSDK_DEP_LIBRARIES ${ZLIB_LIBRARIES})
+    include(FindCURL)
+    list(APPEND AWSSDK_DEP_LIBRARIES ${CURL_LIBRARIES})
+    include(FindOpenSSL)
+    list(APPEND AWSSDK_DEP_LIBRARIES ${OPENSSL_SSL_LIBRARIES} ${OPENSSL_CRYPTO_LIBRARIES})
+  endif()
+  message(STATUS "AWSSDK DEPENDENCIES AWSSDK_DEP_LIBRARIES: ${AWSSDK_DEP_LIBRARIES}")
+
   set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
   set(
@@ -44,9 +63,10 @@ if(BUILD_S3)
   target_link_libraries(
     _torchdata
     PRIVATE
-    ${AWSSDK_LIBRARIES}
     ${Python_LIBRARIES}
     ${ADDITIONAL_ITEMS}
+    ${AWSSDK_LIBRARIES}
+    ${AWSSDK_DEP_LIBRARIES}
     )
 
   set_target_properties(_torchdata PROPERTIES PREFIX "")