Avoid using cudf_helpers.pyx to copy a cuDF DataFrame

.github/workflows/ci_pipe.yml

@@ -137,7 +137,8 @@ jobs:
   test:
     name: Test
     runs-on: linux-amd64-gpu-v100-latest-1
-    timeout-minutes: 60
+    # Consider lowering this back down to 60 minutes per https://github.com/nv-morpheus/Morpheus/issues/1948
+    timeout-minutes: 90
     container:
       credentials:
         username: '$oauthtoken'

python/morpheus/morpheus/_lib/include/morpheus/objects/table_info.hpp

@@ -108,13 +108,13 @@ struct MORPHEUS_EXPORT TableInfoBase
      */
     bool has_sliceable_index() const;
 
+    const std::shared_ptr<const IDataTable>& get_parent() const;
+
   protected:
     TableInfoBase() = default;
 
     TableInfoBase(std::shared_ptr<const IDataTable> parent, TableInfoData data);
 
-    const std::shared_ptr<const IDataTable>& get_parent() const;
-
     TableInfoData& get_data();
     const TableInfoData& get_data() const;
 

python/morpheus/morpheus/_lib/src/messages/meta.cpp

@@ -37,8 +37,9 @@
 #include <pybind11/gil.h>
 #include <pybind11/pybind11.h>
 #include <pybind11/pytypes.h>
-#include <pyerrors.h>  // for PyExc_DeprecationWarning
-#include <warnings.h>  // for PyErr_WarnEx
+#include <pybind11/stl.h>  // IWYU pragma: keep
+#include <pyerrors.h>      // for PyExc_DeprecationWarning
+#include <warnings.h>      // for PyErr_WarnEx
 
 #include <cstddef>  // for size_t
 #include <cstdint>  // for uint8_t
@@ -444,17 +445,24 @@ std::vector<std::string> MessageMetaInterfaceProxy::get_column_names(MessageMeta
 
 py::object MessageMetaInterfaceProxy::get_data_frame(MessageMeta& self)
 {
+    namespace py = pybind11;
     TableInfo info;
 
     {
-        // Need to release the GIL before calling `get_meta()`
-        pybind11::gil_scoped_release no_gil;
+        // Need to release the GIL before calling `get_info()`
+        py::gil_scoped_release no_gil;
 
         // Get the column and convert to cudf
         info = self.get_info();
     }
 
-    return CudfHelper::table_from_table_info(info);
+    auto column_names = info.get_column_names();
+    auto py_df        = info.get_parent()->get_py_object();
+
+    py::gil_scoped_acquire gil;
+    auto py_column_names   = py::cast(column_names);
+    py::object py_df_slice = py_df[py_column_names];
+    return py_df_slice.attr("copy")("deep"_a = true);
 }
 
 py::object MessageMetaInterfaceProxy::df_property(MessageMeta& self)

python/morpheus/morpheus/_lib/tests/messages/test_message_meta.cpp

@@ -26,18 +26,22 @@
 #include "morpheus/objects/tensor.hpp"      // for Tensor
 #include "morpheus/types.hpp"               // for RangeType
 
-#include <gtest/gtest.h>             // for TestInfo, TEST_F
-#include <pybind11/gil.h>            // for gil_scoped_release
+#include <gtest/gtest.h>   // for TestInfo, TEST_F
+#include <pybind11/gil.h>  // for gil_scoped_release
+#include <pybind11/pybind11.h>
+#include <pybind11/pytypes.h>
 #include <rmm/cuda_stream_view.hpp>  // for cuda_stream_per_thread
 #include <rmm/device_buffer.hpp>     // for device_buffer
 
 #include <cstdint>     // for int64_t
 #include <filesystem>  // for operator/, path
 #include <memory>      // for allocator, __shared_ptr_access, shared_ptr, make_shared
+#include <utility>     // for move
 #include <vector>      // for vector
 
 using namespace morpheus;
 using namespace morpheus::test;
+using namespace pybind11::literals;
 
 using TestMessageMeta = morpheus::test::TestMessages;  // NOLINT(readability-identifier-naming)
 
@@ -82,3 +86,25 @@ TEST_F(TestMessageMeta, CopyRangeAndSlicing)
     assert_eq_device_to_host(sliced_meta->get_info().get_column(0), sliced_expected_int);
     assert_eq_device_to_host(sliced_meta->get_info().get_column(1), sliced_expected_double);
 }
+
+TEST_F(TestMessageMeta, Issue1934)
+{
+    // Reproduce issue 1934 (https://github.com/nv-morpheus/Morpheus/issues/1934)
+    // The bug causes a segfault when calling `get_data_frame` on a message meta object
+    namespace py = pybind11;
+    py::gil_scoped_acquire gil;
+
+    auto cudf_mod = py::module_::import("cudf");
+    auto a_col    = py::list();
+    auto v1       = py::list();
+    v1.append(py::str("a"));
+    a_col.attr("append")(std::move(v1));
+    a_col.attr("append")(py::none());
+
+    auto df = cudf_mod.attr("DataFrame")(py::dict("a"_a = std::move(a_col)));
+    df.attr("drop")(0, "inplace"_a = true);
+
+    auto msg = MessageMetaInterfaceProxy::init_python(std::move(df));
+
+    auto df_copy = MessageMetaInterfaceProxy::get_data_frame(*msg);
+}

tests/_utils/test_directories.py

@@ -33,4 +33,3 @@ def __init__(self, tests_dir) -> None:
         self.tests_data_dir = os.path.join(self.tests_dir, 'tests_data')
         self.mock_triton_servers_dir = os.path.join(self.tests_dir, 'mock_triton_server')
         self.mock_rest_server = os.path.join(self.tests_dir, 'mock_rest_server')
-        print(self)