apacheGH-38183: [CI][Python] Use pipx to install GCS testbench (apach…

…e#43852) ### Rationale for this change Installing the GCS testbench using the same Python that's being used to test PyArrow is fragile: some testbench versions may not be compatible, or there could be conflicts among the dependencies of the respective libraries. ### What changes are included in this PR? Use `pipx` to install the GCS testbench in a separate, controlled environment, using an appropriate Python version. ### Are these changes tested? Yes, by CI. ### Are there any user-facing changes? No. * GitHub Issue: apache#38183 Authored-by: Antoine Pitrou <antoine@python.org> Signed-off-by: Sutou Kouhei <kou@clear-code.com>
vibhatha · Aug 29, 2024 · 58415d1 · 58415d1
1 parent 0bc91dd
commit 58415d1
Show file tree

Hide file tree

Showing 16 changed files with 122 additions and 90 deletions.
diff --git a/.github/workflows/cpp.yml b/.github/workflows/cpp.yml
@@ -465,15 +465,17 @@ jobs:
           chmod +x /usr/local/bin/minio.exe
       - name: Set up Python
         uses: actions/setup-python@v5.1.1
+        id: python-install
         with:
           python-version: 3.9
       - name: Install Google Cloud Storage Testbench
-        shell: bash
+        shell: msys2 {0}
+        env:
+          PIPX_BIN_DIR: /usr/local/bin
+          PIPX_PYTHON: ${{ steps.python-install.outputs.python-path }}
         run: |
           ci/scripts/install_gcs_testbench.sh default
-          echo "PYTHON_BIN_DIR=$(cygpath --windows $(dirname $(which python3.exe)))" >> $GITHUB_ENV
       - name: Test
         shell: msys2 {0}
         run: |
-          PATH="$(cygpath --unix ${PYTHON_BIN_DIR}):${PATH}"
           ci/scripts/cpp_test.sh "$(pwd)" "$(pwd)/build"
diff --git a/appveyor.yml b/appveyor.yml
@@ -24,6 +24,7 @@ only_commits:
     - appveyor.yml
     - ci/appveyor*
     - ci/conda*
+    - ci/scripts/*.bat
     - cpp/
     - format/
     - python/

diff --git a/ci/appveyor-cpp-build.bat b/ci/appveyor-cpp-build.bat
@@ -46,7 +46,9 @@ set ARROW_CMAKE_ARGS=-DARROW_DEPENDENCY_SOURCE=CONDA -DARROW_WITH_BZ2=ON
 set ARROW_CXXFLAGS=/WX /MP
 
 @rem Install GCS testbench
+set PIPX_BIN_DIR=C:\Windows\
 call %CD%\ci\scripts\install_gcs_testbench.bat
+storage-testbench -h || exit /B
 
 @rem
 @rem Build and test Arrow C++ libraries (including Parquet)

diff --git a/ci/docker/conda-cpp.dockerfile b/ci/docker/conda-cpp.dockerfile
@@ -42,17 +42,19 @@ RUN mamba install -q -y \
         valgrind && \
     mamba clean --all
 
+# We want to install the GCS testbench using the Conda base environment's Python,
+# because the test environment's Python may later change.
+ENV PIPX_PYTHON=/opt/conda/bin/python3
+COPY ci/scripts/install_gcs_testbench.sh /arrow/ci/scripts
+RUN /arrow/ci/scripts/install_gcs_testbench.sh default
+
 # Ensure npm, node and azurite are on path. npm and node are required to install azurite, which will then need to 
-# be on the path for the tests to run.  
+# be on the path for the tests to run.
 ENV PATH=/opt/conda/envs/arrow/bin:$PATH
 
 COPY ci/scripts/install_azurite.sh /arrow/ci/scripts/
 RUN /arrow/ci/scripts/install_azurite.sh
 
-# We want to install the GCS testbench using the same Python binary that the Conda code will use.
-COPY ci/scripts/install_gcs_testbench.sh /arrow/ci/scripts
-RUN /arrow/ci/scripts/install_gcs_testbench.sh default
-
 COPY ci/scripts/install_sccache.sh /arrow/ci/scripts/
 RUN /arrow/ci/scripts/install_sccache.sh unknown-linux-musl /usr/local/bin
 

diff --git a/ci/docker/conda-python.dockerfile b/ci/docker/conda-python.dockerfile
@@ -32,11 +32,6 @@ RUN mamba install -q -y \
         nomkl && \
     mamba clean --all
 
-# XXX The GCS testbench was already installed in conda-cpp.dockerfile,
-# but we changed the installed Python version above, so we need to reinstall it.
-COPY ci/scripts/install_gcs_testbench.sh /arrow/ci/scripts
-RUN /arrow/ci/scripts/install_gcs_testbench.sh default
-
 ENV ARROW_ACERO=ON \
     ARROW_BUILD_STATIC=OFF \
     ARROW_BUILD_TESTS=OFF \

diff --git a/ci/docker/python-wheel-windows-test-vs2019.dockerfile b/ci/docker/python-wheel-windows-test-vs2019.dockerfile
@@ -35,16 +35,27 @@ RUN setx path "%path%;C:\Program Files\Git\usr\bin"
 RUN wmic product where "name like 'python%%'" call uninstall /nointeractive && \
     rm -rf Python*
 
+# Install the GCS testbench using a well-known Python version.
+# NOTE: cannot use pipx's `--fetch-missing-python` because of
+# https://github.com/pypa/pipx/issues/1521, therefore download Python ourselves.
+RUN choco install -r -y --pre --no-progress python --version=3.11.9
+ENV PIPX_BIN_DIR=C:\\Windows\\
+ENV PIPX_PYTHON="C:\Python311\python.exe"
+COPY ci/scripts/install_gcs_testbench.bat C:/arrow/ci/scripts/
+RUN call "C:\arrow\ci\scripts\install_gcs_testbench.bat" && \
+    storage-testbench -h
+
 # Define the full version number otherwise choco falls back to patch number 0 (3.8 => 3.8.0)
 ARG python=3.8
-RUN (if "%python%"=="3.8" setx PYTHON_VERSION "3.8.10" && setx PATH "%PATH%;C:\Python38;C:\Python38\Scripts") & \
-    (if "%python%"=="3.9" setx PYTHON_VERSION "3.9.13" && setx PATH "%PATH%;C:\Python39;C:\Python39\Scripts") & \
-    (if "%python%"=="3.10" setx PYTHON_VERSION "3.10.11" && setx PATH "%PATH%;C:\Python310;C:\Python310\Scripts") & \
-    (if "%python%"=="3.11" setx PYTHON_VERSION "3.11.9" && setx PATH "%PATH%;C:\Python311;C:\Python311\Scripts") & \
-    (if "%python%"=="3.12" setx PYTHON_VERSION "3.12.4" && setx PATH "%PATH%;C:\Python312;C:\Python312\Scripts") & \
-    (if "%python%"=="3.13" setx PYTHON_VERSION "3.13.0-rc1" && setx PATH "%PATH%;C:\Python313;C:\Python313\Scripts")
+RUN (if "%python%"=="3.8" setx PYTHON_VERSION "3.8.10") & \
+    (if "%python%"=="3.9" setx PYTHON_VERSION "3.9.13") & \
+    (if "%python%"=="3.10" setx PYTHON_VERSION "3.10.11") & \
+    (if "%python%"=="3.11" setx PYTHON_VERSION "3.11.9") & \
+    (if "%python%"=="3.12" setx PYTHON_VERSION "3.12.4") & \
+    (if "%python%"=="3.13" setx PYTHON_VERSION "3.13.0-rc1")
 
 # Install archiver to extract xz archives
-RUN choco install -r -y --pre --no-progress python --version=%PYTHON_VERSION% & \
-    python -m pip install --no-cache-dir -U pip setuptools & \
+RUN choco install -r -y --pre --no-progress --force python --version=%PYTHON_VERSION% && \
     choco install --no-progress -r -y archiver
+
+ENV PYTHON=$python
diff --git a/ci/docker/ubuntu-20.04-cpp-minimal.dockerfile b/ci/docker/ubuntu-20.04-cpp-minimal.dockerfile
@@ -33,6 +33,7 @@ RUN apt-get update -y -q && \
         libssl-dev \
         libcurl4-openssl-dev \
         python3-pip \
+        python3-venv \
         tzdata \
         wget && \
     apt-get clean && \

diff --git a/ci/docker/ubuntu-22.04-cpp-minimal.dockerfile b/ci/docker/ubuntu-22.04-cpp-minimal.dockerfile
@@ -33,6 +33,7 @@ RUN apt-get update -y -q && \
         libssl-dev \
         libcurl4-openssl-dev \
         python3-pip \
+        python3-venv \
         tzdata \
         wget && \
     apt-get clean && \

diff --git a/ci/docker/ubuntu-24.04-cpp-minimal.dockerfile b/ci/docker/ubuntu-24.04-cpp-minimal.dockerfile
@@ -33,6 +33,7 @@ RUN apt-get update -y -q && \
         libssl-dev \
         libcurl4-openssl-dev \
         python3-pip \
+        python3-venv \
         tzdata \
         tzdata-legacy \
         wget && \

diff --git a/ci/scripts/install_gcs_testbench.bat b/ci/scripts/install_gcs_testbench.bat
@@ -17,9 +17,18 @@
 
 @echo on
 
-set GCS_TESTBENCH_VERSION="v0.36.0"
+set GCS_TESTBENCH_VERSION="v0.40.0"
+
+set PIPX_FLAGS=--verbose
+if NOT "%PIPX_PYTHON%"=="" (
+  set PIPX_FLAGS=--python %PIPX_PYTHON% %PIPX_FLAGS%
+)
+
+python -m pip install -U pipx || exit /B 1
 
 @REM Install GCS testbench %GCS_TESTBENCH_VERSION%
-python -m pip install  ^
+pipx install %PIPX_FLAGS% ^
         "https://github.com/googleapis/storage-testbench/archive/%GCS_TESTBENCH_VERSION%.tar.gz" ^
         || exit /B 1
+
+pipx list --verbose
diff --git a/ci/scripts/install_gcs_testbench.sh b/ci/scripts/install_gcs_testbench.sh
@@ -17,7 +17,7 @@
 # specific language governing permissions and limitations
 # under the License.
 
-set -e
+set -ex
 
 if [ "$#" -ne 1 ]; then
   echo "Usage: $0 <storage-testbench version>"
@@ -34,19 +34,23 @@ case "$(uname -m)" in
     ;;
 esac
 
-# On newer pythons install into the system will fail, so override that
-export PIP_BREAK_SYSTEM_PACKAGES=1
-
 version=$1
 if [[ "${version}" -eq "default" ]]; then
   version="v0.39.0"
-  # Latests versions of Testbench require newer setuptools
-  python3 -m pip install --upgrade setuptools
 fi
 
+: ${PIPX_PYTHON:=$(which python3)}
+
+export PIP_BREAK_SYSTEM_PACKAGES=1
+${PIPX_PYTHON} -m pip install -U pipx
+
 # This script is run with PYTHON undefined in some places,
 # but those only use older pythons.
 if [[ -z "${PYTHON_VERSION}" ]] || [[ "${PYTHON_VERSION}" != "3.13" ]]; then
-  python3 -m pip install \
-    "https://github.com/googleapis/storage-testbench/archive/${version}.tar.gz"
+  pipx_flags=--verbose
+  if [[ $(id -un) == "root" ]]; then
+    # Install globally as /root/.local/bin is typically not in $PATH
+    pipx_flags="${pipx_flags} --global"
+  fi
+  ${PIPX_PYTHON} -m pipx install ${pipx_flags} "https://github.com/googleapis/storage-testbench/archive/${version}.tar.gz"
 fi
diff --git a/ci/scripts/python_wheel_windows_test.bat b/ci/scripts/python_wheel_windows_test.bat
@@ -37,28 +37,32 @@ set PYARROW_TEST_TENSORFLOW=ON
 set ARROW_TEST_DATA=C:\arrow\testing\data
 set PARQUET_TEST_DATA=C:\arrow\cpp\submodules\parquet-testing\data
 
-@REM Install testing dependencies
-pip install -r C:\arrow\python\requirements-wheel-test.txt || exit /B 1
+@REM List installed Pythons
+py -0p
+
+set PYTHON_CMD=py -%PYTHON%
 
-@REM Install GCS testbench
-call "C:\arrow\ci\scripts\install_gcs_testbench.bat"
+%PYTHON_CMD% -m pip install -U pip setuptools || exit /B 1
+
+@REM Install testing dependencies
+%PYTHON_CMD% -m pip install -r C:\arrow\python\requirements-wheel-test.txt || exit /B 1
 
 @REM Install the built wheels
-python -m pip install --no-index --find-links=C:\arrow\python\dist\ pyarrow || exit /B 1 
+%PYTHON_CMD% -m pip install --no-index --find-links=C:\arrow\python\dist\ pyarrow || exit /B 1
 
 @REM Test that the modules are importable
-python -c "import pyarrow" || exit /B 1
-python -c "import pyarrow._gcsfs" || exit /B 1
-python -c "import pyarrow._hdfs" || exit /B 1 
-python -c "import pyarrow._s3fs" || exit /B 1
-python -c "import pyarrow.csv" || exit /B 1
-python -c "import pyarrow.dataset" || exit /B 1
-python -c "import pyarrow.flight" || exit /B 1
-python -c "import pyarrow.fs" || exit /B 1
-python -c "import pyarrow.json" || exit /B 1
-python -c "import pyarrow.orc" || exit /B 1
-python -c "import pyarrow.parquet" || exit /B 1
-python -c "import pyarrow.substrait" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow._gcsfs" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow._hdfs" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow._s3fs" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow.csv" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow.dataset" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow.flight" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow.fs" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow.json" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow.orc" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow.parquet" || exit /B 1
+%PYTHON_CMD% -c "import pyarrow.substrait" || exit /B 1
 
 @rem Download IANA Timezone Database for ORC C++
 curl https://cygwin.osuosl.org/noarch/release/tzdata/tzdata-2024a-1.tar.xz --output tzdata.tar.xz || exit /B
@@ -67,4 +71,4 @@ arc unarchive tzdata.tar.xz %USERPROFILE%\Downloads\test\tzdata
 set TZDIR=%USERPROFILE%\Downloads\test\tzdata\usr\share\zoneinfo
 
 @REM Execute unittest
-pytest -r s --pyargs pyarrow || exit /B 1
+%PYTHON_CMD% -m pytest -r s --pyargs pyarrow || exit /B 1
diff --git a/cpp/src/arrow/filesystem/gcsfs_test.cc b/cpp/src/arrow/filesystem/gcsfs_test.cc
@@ -95,52 +95,52 @@ class GcsTestbench : public ::testing::Environment {
     if (const auto* env = std::getenv("PYTHON")) {
       names = {env};
     }
-    auto error = std::string(
-        "Could not start GCS emulator."
-        " Used the following list of python interpreter names:");
-    for (const auto& interpreter : names) {
-      auto exe_path = bp::search_path(interpreter);
-      error += " " + interpreter;
-      if (exe_path.empty()) {
-        error += " (exe not found)";
-        continue;
-      }
+    auto error = std::string("Could not start GCS emulator 'storage-testbench'");
 
-      bp::ipstream output;
-      server_process_ = bp::child(exe_path, "-m", "testbench", "--port", port_, group_,
-                                  bp::std_err > output);
+    auto testbench_is_running = [](bp::child& process, bp::ipstream& output) {
       // Wait for message: "* Restarting with"
-      auto testbench_is_running = [&output, this](bp::child& process) {
-        std::string line;
-        std::chrono::time_point<std::chrono::steady_clock> end =
-            std::chrono::steady_clock::now() + std::chrono::seconds(10);
-        while (server_process_.valid() && server_process_.running() &&
-               std::chrono::steady_clock::now() < end) {
-          if (output.peek() && std::getline(output, line)) {
-            std::cerr << line << std::endl;
-            if (line.find("* Restarting with") != std::string::npos) return true;
-          } else {
-            std::this_thread::sleep_for(std::chrono::milliseconds(20));
-          }
+      std::string line;
+      std::chrono::time_point<std::chrono::steady_clock> end =
+          std::chrono::steady_clock::now() + std::chrono::seconds(10);
+      while (process.valid() && process.running() &&
+             std::chrono::steady_clock::now() < end) {
+        if (output.peek() && std::getline(output, line)) {
+          std::cerr << line << std::endl;
+          if (line.find("* Restarting with") != std::string::npos) return true;
+        } else {
+          std::this_thread::sleep_for(std::chrono::milliseconds(20));
         }
-        return false;
-      };
+      }
+      return false;
+    };
 
-      if (testbench_is_running(server_process_)) break;
-      error += " (failed to start)";
-      server_process_.terminate();
-      server_process_.wait();
+    auto exe_path = bp::search_path("storage-testbench");
+    if (!exe_path.empty()) {
+      bp::ipstream output;
+      server_process_ =
+          bp::child(exe_path, "--port", port_, group_, bp::std_err > output);
+      if (!testbench_is_running(server_process_, output)) {
+        error += " (failed to start)";
+        server_process_.terminate();
+        server_process_.wait();
+      }
+    } else {
+      error += " (exe not found)";
+    }
+    if (!server_process_.valid()) {
+      error_ = std::move(error);
     }
-    if (server_process_.valid() && server_process_.valid()) return;
-    error_ = std::move(error);
   }
 
   bool running() { return server_process_.running(); }
 
   ~GcsTestbench() override {
     // Brutal shutdown, kill the full process group because the GCS testbench may launch
     // additional children.
-    group_.terminate();
+    try {
+      group_.terminate();
+    } catch (bp::process_error&) {
+    }
     if (server_process_.valid()) {
       server_process_.wait();
     }

diff --git a/python/pyarrow/tests/conftest.py b/python/pyarrow/tests/conftest.py
@@ -233,17 +233,16 @@ def minio_server_health_check(address):
 def gcs_server():
     port = find_free_port()
     env = os.environ.copy()
-    args = [sys.executable, '-m', 'testbench', '--port', str(port)]
+    exe = 'storage-testbench'
+    args = [exe, '--port', str(port)]
     proc = None
     try:
-        # check first if testbench module is available
-        import testbench  # noqa:F401
         # start server
         proc = subprocess.Popen(args, env=env)
         # Make sure the server is alive.
         if proc.poll() is not None:
             pytest.skip(f"Command {args} did not start server successfully!")
-    except (ModuleNotFoundError, OSError) as e:
+    except OSError as e:
         pytest.skip(f"Command {args} failed to execute: {e}")
     else:
         yield {

diff --git a/python/scripts/run_emscripten_tests.py b/python/scripts/run_emscripten_tests.py
@@ -335,7 +335,7 @@ def _load_pyarrow_in_runner(driver, wheel_name):
         """
 import pyarrow,pathlib
 pyarrow_dir = pathlib.Path(pyarrow.__file__).parent
-pytest.main([pyarrow_dir, '-v'])
+pytest.main([pyarrow_dir, '-r', 's'])
 """,
         wait_for_terminate=False,
     )

diff --git a/r/tests/testthat/test-gcs.R b/r/tests/testthat/test-gcs.R
@@ -116,12 +116,12 @@ test_that("GcsFileSystem$create() can read json_credentials", {
 })
 
 skip_on_cran()
-skip_if_not(system('python -c "import testbench"') == 0, message = "googleapis-storage-testbench is not installed.")
+skip_if_not(system("storage-testbench -h") == 0, message = "googleapis-storage-testbench is not installed.")
 library(dplyr)
 
 testbench_port <- Sys.getenv("TESTBENCH_PORT", "9001")
 
-pid_minio <- sys::exec_background("python", c("-m", "testbench", "--port", testbench_port),
+pid_minio <- sys::exec_background("storage-testbench", c("--port", testbench_port),
   std_out = FALSE,
   std_err = FALSE # TODO: is there a good place to send output?
 )