Merge pull request #17379 from bdice/branch-25.02-merge-24.12

Forward-merge branch-24.12 to branch-25.02

.github/workflows/pr.yaml

@@ -10,6 +10,7 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
+  # Please keep pr-builder as the top job here
   pr-builder:
     needs:
       - changed-files
@@ -37,13 +38,23 @@ jobs:
       - unit-tests-cudf-pandas
       - pandas-tests
       - pandas-tests-diff
+      - telemetry-setup
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/[email protected]
     if: always()
     with:
       needs: ${{ toJSON(needs) }}
+  telemetry-setup:
+    continue-on-error: true
+    runs-on: ubuntu-latest
+    env:
+      OTEL_SERVICE_NAME: 'pr-cudf'
+    steps:
+      - name: Telemetry setup
+        uses: rapidsai/shared-actions/telemetry-dispatch-stash-base-env-vars@main
   changed-files:
     secrets: inherit
+    needs: telemetry-setup
     uses: rapidsai/shared-workflows/.github/workflows/[email protected]
     with:
       files_yaml: |
@@ -91,9 +102,11 @@ jobs:
           - '!notebooks/**'
   checks:
     secrets: inherit
+    needs: telemetry-setup
     uses: rapidsai/shared-workflows/.github/workflows/[email protected]
     with:
       enable_check_generated_files: false
+      ignored_pr_jobs: "telemetry-summarize"
   conda-cpp-build:
     needs: checks
     secrets: inherit
@@ -259,6 +272,7 @@ jobs:
       script: ci/test_wheel_dask_cudf.sh
   devcontainer:
     secrets: inherit
+    needs: telemetry-setup
     uses: rapidsai/shared-workflows/.github/workflows/[email protected]
     with:
       arch: '["amd64"]'
@@ -298,3 +312,18 @@ jobs:
         node_type: cpu4
         build_type: pull-request
         run_script: "ci/cudf_pandas_scripts/pandas-tests/diff.sh"
+
+  telemetry-summarize:
+    runs-on: ubuntu-latest
+    needs: pr-builder
+    if: always()
+    continue-on-error: true
+    steps:
+      - name: Load stashed telemetry env vars
+        uses: rapidsai/shared-actions/telemetry-dispatch-load-base-env-vars@main
+        with:
+            load_service_name: true
+      - name: Telemetry summarize
+        uses: rapidsai/shared-actions/telemetry-dispatch-write-summary@main
+        with:
+          cert_concat: "${{ secrets.OTEL_EXPORTER_OTLP_CA_CERTIFICATE }};${{ secrets.OTEL_EXPORTER_OTLP_CLIENT_CERTIFICATE }};${{ secrets.OTEL_EXPORTER_OTLP_CLIENT_KEY }}"

.github/workflows/test.yaml

@@ -121,8 +121,6 @@ jobs:
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/[email protected]
     with:
-      # This selects "ARCH=amd64 + the latest supported Python + CUDA".
-      matrix_filter: map(select(.ARCH == "amd64")) | group_by(.CUDA_VER|split(".")|map(tonumber)|.[0]) | map(max_by([(.PY_VER|split(".")|map(tonumber)), (.CUDA_VER|split(".")|map(tonumber))]))
       build_type: nightly
       branch: ${{ inputs.branch }}
       date: ${{ inputs.date }}
@@ -153,8 +151,6 @@ jobs:
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/[email protected]
     with:
-      # This selects "ARCH=amd64 + the latest supported Python + CUDA".
-      matrix_filter: map(select(.ARCH == "amd64")) | group_by(.CUDA_VER|split(".")|map(tonumber)|.[0]) | map(max_by([(.PY_VER|split(".")|map(tonumber)), (.CUDA_VER|split(".")|map(tonumber))]))
       build_type: nightly
       branch: ${{ inputs.branch }}
       date: ${{ inputs.date }}
@@ -164,8 +160,6 @@ jobs:
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/[email protected]
     with:
-      # This selects "ARCH=amd64 + the latest supported Python + CUDA".
-      matrix_filter: map(select(.ARCH == "amd64")) | group_by(.CUDA_VER|split(".")|map(tonumber)|.[0]) | map(max_by([(.PY_VER|split(".")|map(tonumber)), (.CUDA_VER|split(".")|map(tonumber))]))
       build_type: nightly
       branch: ${{ inputs.branch }}
       date: ${{ inputs.date }}

ci/run_cudf_polars_polars_tests.sh

@@ -12,8 +12,29 @@ DESELECTED_TESTS=(
     "tests/unit/streaming/test_streaming_sort.py::test_streaming_sort[True]" # relies on polars built in debug mode
     "tests/unit/test_cpu_check.py::test_check_cpu_flags_skipped_no_flags" # Mock library error
     "tests/docs/test_user_guide.py" # No dot binary in CI image
+    "tests/unit/test_polars_import.py::test_fork_safety" # test started to hang in polars-1.14
+    "tests/unit/operations/test_join.py::test_join_4_columns_with_validity" # fails in some systems, see https://github.com/pola-rs/polars/issues/19870
+    "tests/unit/io/test_csv.py::test_read_web_file" # fails in rockylinux8 due to SSL CA issues
 )
 
+if [[ $(arch) == "aarch64" ]]; then
+    # The binary used for TPC-H generation is compiled for x86_64, not aarch64.
+    DESELECTED_TESTS+=("tests/benchmark/test_pdsh.py::test_pdsh")
+    # The connectorx package is not available on arm
+    DESELECTED_TESTS+=("tests/unit/io/database/test_read.py::test_read_database")
+    # The necessary timezone information cannot be found in our CI image.
+    DESELECTED_TESTS+=("tests/unit/io/test_parquet.py::test_parametric_small_page_mask_filtering")
+    DESELECTED_TESTS+=("tests/unit/testing/test_assert_series_equal.py::test_assert_series_equal_parametric")
+    DESELECTED_TESTS+=("tests/unit/operations/test_join.py::test_join_4_columns_with_validity")
+else
+    # Ensure that we don't run dbgen when it uses newer symbols than supported by the glibc version in the CI image.
+    glibc_minor_version=$(ldd --version | head -1 | grep -o "[0-9]\.[0-9]\+" | tail -1 | cut -d '.' -f2)
+    latest_glibc_symbol_found=$(nm py-polars/tests/benchmark/data/pdsh/dbgen/dbgen | grep GLIBC | grep -o "[0-9]\.[0-9]\+" | sort --version-sort | tail -1 | cut -d "." -f 2)
+    if [[ ${glibc_minor_version} -lt ${latest_glibc_symbol_found} ]]; then
+        DESELECTED_TESTS+=("tests/benchmark/test_pdsh.py::test_pdsh")
+    fi
+fi
+
 DESELECTED_TESTS=$(printf -- " --deselect %s" "${DESELECTED_TESTS[@]}")
 python -m pytest \
        --import-mode=importlib \

conda/environments/all_cuda-118_arch-x86_64.yaml

@@ -66,7 +66,7 @@ dependencies:
 - pandas
 - pandas>=2.0,<2.2.4dev0
 - pandoc
-- polars>=1.11,<1.14
+- polars>=1.11,<1.15
 - pre-commit
 - ptxcompiler
 - pyarrow>=14.0.0,<19.0.0a0

conda/environments/all_cuda-125_arch-x86_64.yaml

@@ -64,7 +64,7 @@ dependencies:
 - pandas
 - pandas>=2.0,<2.2.4dev0
 - pandoc
-- polars>=1.11,<1.14
+- polars>=1.11,<1.15
 - pre-commit
 - pyarrow>=14.0.0,<19.0.0a0
 - pydata-sphinx-theme!=0.14.2

conda/recipes/cudf-polars/meta.yaml

@@ -43,7 +43,7 @@ requirements:
   run:
     - python
     - pylibcudf ={{ version }}
-    - polars >=1.11,<1.14
+    - polars >=1.11,<1.15
     - {{ pin_compatible('cuda-version', max_pin='x', min_pin='x') }}
 
 test:

cpp/CMakeLists.txt

@@ -464,6 +464,7 @@ add_library(
   src/io/avro/avro_gpu.cu
   src/io/avro/reader_impl.cu
   src/io/comp/brotli_dict.cpp
+  src/io/comp/comp.cpp
   src/io/comp/cpu_unbz2.cpp
   src/io/comp/debrotli.cu
   src/io/comp/gpuinflate.cu

cpp/benchmarks/CMakeLists.txt

@@ -354,10 +354,7 @@ ConfigureNVBench(
 
 # ##################################################################################################
 # * strings benchmark -------------------------------------------------------------------
-ConfigureBench(
-  STRINGS_BENCH string/factory.cu string/repeat_strings.cpp string/replace.cpp string/translate.cpp
-  string/url_decode.cu
-)
+ConfigureBench(STRINGS_BENCH string/factory.cu string/repeat_strings.cpp string/url_decode.cu)
 
 ConfigureNVBench(
   STRINGS_NVBENCH
@@ -381,11 +378,13 @@ ConfigureNVBench(
   string/lengths.cpp
   string/like.cpp
   string/make_strings_column.cu
+  string/replace.cpp
   string/replace_re.cpp
   string/reverse.cpp
   string/slice.cpp
   string/split.cpp
   string/split_re.cpp
+  string/translate.cpp
 )
 
 # ##################################################################################################

cpp/benchmarks/io/json/json_reader_input.cpp

@@ -24,17 +24,19 @@
 
 #include <nvbench/nvbench.cuh>
 
-// Size of the data in the benchmark dataframe; chosen to be low enough to allow benchmarks to
-// run on most GPUs, but large enough to allow highest throughput
-constexpr size_t data_size         = 512 << 20;
+// Default size of the data in the benchmark dataframe; chosen to be low enough to allow benchmarks
+// to run on most GPUs, but large enough to allow highest throughput
+constexpr size_t default_data_size = 512 << 20;
 constexpr cudf::size_type num_cols = 64;
 
 void json_read_common(cuio_source_sink_pair& source_sink,
                       cudf::size_type num_rows_to_read,
-                      nvbench::state& state)
+                      nvbench::state& state,
+                      cudf::io::compression_type comptype = cudf::io::compression_type::NONE,
+                      size_t data_size                    = default_data_size)
 {
   cudf::io::json_reader_options read_opts =
-    cudf::io::json_reader_options::builder(source_sink.make_source_info());
+    cudf::io::json_reader_options::builder(source_sink.make_source_info()).compression(comptype);
 
   auto mem_stats_logger = cudf::memory_stats_logger();
   state.set_cuda_stream(nvbench::make_cuda_stream_view(cudf::get_default_stream().value()));
@@ -57,15 +59,21 @@ void json_read_common(cuio_source_sink_pair& source_sink,
   state.add_buffer_size(source_sink.size(), "encoded_file_size", "encoded_file_size");
 }
 
-cudf::size_type json_write_bm_data(cudf::io::sink_info sink,
-                                   std::vector<cudf::type_id> const& dtypes)
+cudf::size_type json_write_bm_data(
+  cudf::io::sink_info sink,
+  std::vector<cudf::type_id> const& dtypes,
+  cudf::io::compression_type comptype = cudf::io::compression_type::NONE,
+  size_t data_size                    = default_data_size)
 {
   auto const tbl = create_random_table(
     cycle_dtypes(dtypes, num_cols), table_size_bytes{data_size}, data_profile_builder());
   auto const view = tbl->view();
 
   cudf::io::json_writer_options const write_opts =
-    cudf::io::json_writer_options::builder(sink, view).na_rep("null").rows_per_chunk(100'000);
+    cudf::io::json_writer_options::builder(sink, view)
+      .na_rep("null")
+      .rows_per_chunk(100'000)
+      .compression(comptype);
   cudf::io::write_json(write_opts);
   return view.num_rows();
 }
@@ -87,6 +95,26 @@ void BM_json_read_io(nvbench::state& state, nvbench::type_list<nvbench::enum_typ
   json_read_common(source_sink, num_rows, state);
 }
 
+template <cudf::io::compression_type comptype, io_type IO>
+void BM_json_read_compressed_io(
+  nvbench::state& state, nvbench::type_list<nvbench::enum_type<comptype>, nvbench::enum_type<IO>>)
+{
+  size_t const data_size = state.get_int64("data_size");
+  cuio_source_sink_pair source_sink(IO);
+  auto const d_type = get_type_or_group({static_cast<int32_t>(data_type::INTEGRAL),
+                                         static_cast<int32_t>(data_type::FLOAT),
+                                         static_cast<int32_t>(data_type::DECIMAL),
+                                         static_cast<int32_t>(data_type::TIMESTAMP),
+                                         static_cast<int32_t>(data_type::DURATION),
+                                         static_cast<int32_t>(data_type::STRING),
+                                         static_cast<int32_t>(data_type::LIST),
+                                         static_cast<int32_t>(data_type::STRUCT)});
+  auto const num_rows =
+    json_write_bm_data(source_sink.make_sink_info(), d_type, comptype, data_size);
+
+  json_read_common(source_sink, num_rows, state, comptype, data_size);
+}
+
 template <data_type DataType, io_type IO>
 void BM_json_read_data_type(
   nvbench::state& state, nvbench::type_list<nvbench::enum_type<DataType>, nvbench::enum_type<IO>>)
@@ -110,8 +138,9 @@ using d_type_list = nvbench::enum_type_list<data_type::INTEGRAL,
 using io_list =
   nvbench::enum_type_list<io_type::FILEPATH, io_type::HOST_BUFFER, io_type::DEVICE_BUFFER>;
 
-using compression_list =
-  nvbench::enum_type_list<cudf::io::compression_type::SNAPPY, cudf::io::compression_type::NONE>;
+using compression_list = nvbench::enum_type_list<cudf::io::compression_type::GZIP,
+                                                 cudf::io::compression_type::SNAPPY,
+                                                 cudf::io::compression_type::NONE>;
 
 NVBENCH_BENCH_TYPES(BM_json_read_data_type,
                     NVBENCH_TYPE_AXES(d_type_list, nvbench::enum_type_list<io_type::DEVICE_BUFFER>))
@@ -123,3 +152,10 @@ NVBENCH_BENCH_TYPES(BM_json_read_io, NVBENCH_TYPE_AXES(io_list))
   .set_name("json_read_io")
   .set_type_axes_names({"io"})
   .set_min_samples(4);
+
+NVBENCH_BENCH_TYPES(BM_json_read_compressed_io,
+                    NVBENCH_TYPE_AXES(compression_list, nvbench::enum_type_list<io_type::FILEPATH>))
+  .set_name("json_read_compressed_io")
+  .set_type_axes_names({"compression_type", "io"})
+  .add_int64_power_of_two_axis("data_size", nvbench::range(20, 29, 1))
+  .set_min_samples(4);

cpp/benchmarks/io/nvbench_helpers.hpp

@@ -76,6 +76,7 @@ NVBENCH_DECLARE_ENUM_TYPE_STRINGS(
   [](auto value) {
     switch (value) {
       case cudf::io::compression_type::SNAPPY: return "SNAPPY";
+      case cudf::io::compression_type::GZIP: return "GZIP";
       case cudf::io::compression_type::NONE: return "NONE";
       default: return "Unknown";
     }