Merge pull request #10101 from rapidsai/branch-22.02

rapidsai · Feb 2, 2022 · f39f559 · f39f559
2 parents 06540b9 + a7d88cd
commit f39f559
Show file tree

Hide file tree

Showing 679 changed files with 27,842 additions and 13,014 deletions.
diff --git a/.clang-tidy b/.clang-tidy
@@ -0,0 +1,27 @@
+---
+Checks:    
+      'modernize-*,
+       -modernize-use-equals-default,
+       -modernize-concat-nested-namespaces,
+       -modernize-use-trailing-return-type'
+
+      # -modernize-use-equals-default        # auto-fix is broken (doesn't insert =default correctly)
+      # -modernize-concat-nested-namespaces  # auto-fix is broken (can delete code)
+      # -modernize-use-trailing-return-type  # just a preference
+
+WarningsAsErrors: ''
+HeaderFilterRegex: ''
+AnalyzeTemporaryDtors: false
+FormatStyle:     none
+CheckOptions:
+ - key:             modernize-loop-convert.MaxCopySize
+   value:           '16'
+ - key:             modernize-loop-convert.MinConfidence
+   value:           reasonable
+ - key:             modernize-pass-by-value.IncludeStyle
+   value:           llvm
+ - key:             modernize-replace-auto-ptr.IncludeStyle
+   value:           llvm
+ - key:             modernize-use-nullptr.NullMacros
+   value:           'NULL'
+...
diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
@@ -72,7 +72,7 @@ repos:
                 args: ['-fallback-style=none']
               - id: cmake-format
                 name: cmake-format
-                entry: bash cpp/scripts/run-cmake-format.sh cmake-format
+                entry: ./cpp/scripts/run-cmake-format.sh cmake-format
                 language: python
                 types: [cmake]
                 # Note that pre-commit autoupdate does not update the versions
@@ -81,7 +81,7 @@ repos:
                   - cmake-format==0.6.11
               - id: cmake-lint
                 name: cmake-lint
-                entry: bash cpp/scripts/run-cmake-format.sh cmake-lint
+                entry: ./cpp/scripts/run-cmake-format.sh cmake-lint
                 language: python
                 types: [cmake]
                 # Note that pre-commit autoupdate does not update the versions

diff --git a/CHANGELOG.md b/CHANGELOG.md
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
@@ -86,7 +86,7 @@ git submodule update --init --remote --recursive
 ```bash
 # create the conda environment (assuming in base `cudf` directory)
 # note: RAPIDS currently doesn't support `channel_priority: strict`; use `channel_priority: flexible` instead
-conda env create --name cudf_dev --file conda/environments/cudf_dev_cuda11.0.yml
+conda env create --name cudf_dev --file conda/environments/cudf_dev_cuda11.5.yml
 # activate the environment
 conda activate cudf_dev
 ```

diff --git a/build.sh b/build.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 
-# Copyright (c) 2019-2021, NVIDIA CORPORATION.
+# Copyright (c) 2019-2022, NVIDIA CORPORATION.
 
 # cuDF build script
 
@@ -17,7 +17,7 @@ ARGS=$*
 # script, and that this script resides in the repo dir!
 REPODIR=$(cd $(dirname $0); pwd)
 
-VALIDARGS="clean libcudf cudf dask_cudf benchmarks tests libcudf_kafka cudf_kafka custreamz -v -g -n -l --allgpuarch --disable_nvtx --show_depr_warn --ptds -h"
+VALIDARGS="clean libcudf cudf dask_cudf benchmarks tests libcudf_kafka cudf_kafka custreamz -v -g -n -l --allgpuarch --disable_nvtx --show_depr_warn --ptds -h --build_metrics --incl_cache_stats"
 HELP="$0 [clean] [libcudf] [cudf] [dask_cudf] [benchmarks] [tests] [libcudf_kafka] [cudf_kafka] [custreamz] [-v] [-g] [-n] [-h] [-l] [--cmake-args=\\\"<args>\\\"]
    clean                         - remove all existing build artifacts and configuration (start
                                    over)
@@ -37,6 +37,8 @@ HELP="$0 [clean] [libcudf] [cudf] [dask_cudf] [benchmarks] [tests] [libcudf_kafk
    --disable_nvtx                - disable inserting NVTX profiling ranges
    --show_depr_warn              - show cmake deprecation warnings
    --ptds                        - enable per-thread default stream
+   --build_metrics               - generate build metrics report for libcudf
+   --incl_cache_stats            - include cache statistics in build metrics report
    --cmake-args=\\\"<args>\\\"   - pass arbitrary list of CMake configuration options (escape all quotes in argument)
    -h | --h[elp]                 - print this text
 
@@ -61,6 +63,8 @@ BUILD_NVTX=ON
 BUILD_TESTS=OFF
 BUILD_DISABLE_DEPRECATION_WARNING=ON
 BUILD_PER_THREAD_DEFAULT_STREAM=OFF
+BUILD_REPORT_METRICS=OFF
+BUILD_REPORT_INCL_CACHE_STATS=OFF
 
 # Set defaults for vars that may not have been defined externally
 #  FIXME: if INSTALL_PREFIX is not set, check PREFIX, then check
@@ -144,6 +148,14 @@ fi
 if hasArg --ptds; then
     BUILD_PER_THREAD_DEFAULT_STREAM=ON
 fi
+if hasArg --build_metrics; then
+    BUILD_REPORT_METRICS=ON
+fi
+
+if hasArg --incl_cache_stats; then
+    BUILD_REPORT_INCL_CACHE_STATS=ON
+fi
+
 
 # If clean given, run it prior to any other steps
 if hasArg clean; then
@@ -172,6 +184,15 @@ if buildAll || hasArg libcudf; then
         echo "Building for *ALL* supported GPU architectures..."
     fi
 
+    # get the current count before the compile starts
+    FILES_IN_CCACHE=""
+    if [[ "$BUILD_REPORT_INCL_CACHE_STATS" == "ON" && -x "$(command -v ccache)" ]]; then
+        FILES_IN_CCACHE=$(ccache -s | grep "files in cache")
+        echo "$FILES_IN_CCACHE"
+        # zero the ccache statistics
+        ccache -z
+    fi
+
     cmake -S $REPODIR/cpp -B ${LIB_BUILD_DIR} \
           -DCMAKE_INSTALL_PREFIX=${INSTALL_PREFIX} \
           ${CUDF_CMAKE_CUDA_ARCHITECTURES} \
@@ -185,7 +206,31 @@ if buildAll || hasArg libcudf; then
 
     cd ${LIB_BUILD_DIR}
 
+    compile_start=$(date +%s)
     cmake --build . -j${PARALLEL_LEVEL} ${VERBOSE_FLAG}
+    compile_end=$(date +%s)
+    compile_total=$(( compile_end - compile_start ))
+
+    # Record build times
+    if [[ "$BUILD_REPORT_METRICS" == "ON" && -f "${LIB_BUILD_DIR}/.ninja_log" ]]; then
+        echo "Formatting build metrics"
+        python ${REPODIR}/cpp/scripts/sort_ninja_log.py ${LIB_BUILD_DIR}/.ninja_log --fmt xml > ${LIB_BUILD_DIR}/ninja_log.xml
+        MSG="<p>"
+        # get some ccache stats after the compile
+        if [[ "$BUILD_REPORT_INCL_CACHE_STATS"=="ON" && -x "$(command -v ccache)" ]]; then
+           MSG="${MSG}<br/>$FILES_IN_CCACHE"
+           HIT_RATE=$(ccache -s | grep "cache hit rate")
+           MSG="${MSG}<br/>${HIT_RATE}"
+        fi
+        MSG="${MSG}<br/>parallel setting: $PARALLEL_LEVEL"
+        MSG="${MSG}<br/>parallel build time: $compile_total seconds"
+        if [[ -f "${LIB_BUILD_DIR}/libcudf.so" ]]; then
+           LIBCUDF_FS=$(ls -lh ${LIB_BUILD_DIR}/libcudf.so | awk '{print $5}')
+           MSG="${MSG}<br/>libcudf.so size: $LIBCUDF_FS"
+        fi
+        echo "$MSG"
+        python ${REPODIR}/cpp/scripts/sort_ninja_log.py ${LIB_BUILD_DIR}/.ninja_log --fmt html --msg "$MSG" > ${LIB_BUILD_DIR}/ninja_log.html
+    fi
 
     if [[ ${INSTALL_TARGET} != "" ]]; then
         cmake --build . -j${PARALLEL_LEVEL} --target install ${VERBOSE_FLAG}

diff --git a/ci/benchmark/build.sh b/ci/benchmark/build.sh
@@ -37,7 +37,7 @@ export GBENCH_BENCHMARKS_DIR="$WORKSPACE/cpp/build/gbenchmarks/"
 export LIBCUDF_KERNEL_CACHE_PATH="$HOME/.jitify-cache"
 
 # Dask & Distributed git tag
-export DASK_DISTRIBUTED_GIT_TAG='2021.11.2'
+export DASK_DISTRIBUTED_GIT_TAG='2022.01.0'
 
 function remove_libcudf_kernel_cache_dir {
     EXITCODE=$?
@@ -98,7 +98,7 @@ conda list --show-channel-urls
 ################################################################################
 
 logger "Build libcudf..."
-if [[ ${BUILD_MODE} == "pull-request" ]]; then
+if [[ "${BUILD_MODE}" == "pull-request" ]]; then
     "$WORKSPACE/build.sh" clean libcudf cudf dask_cudf benchmarks tests --ptds
 else
     "$WORKSPACE/build.sh" clean libcudf cudf dask_cudf benchmarks tests -l --ptds

diff --git a/ci/cpu/build.sh b/ci/cpu/build.sh
@@ -78,6 +78,14 @@ if [ "$BUILD_LIBCUDF" == '1' ]; then
   mkdir -p ${CONDA_BLD_DIR}/libcudf/work
   cp -r ${CONDA_BLD_DIR}/work/* ${CONDA_BLD_DIR}/libcudf/work
 
+  # Copy libcudf build metrics results
+  LIBCUDF_BUILD_DIR=$CONDA_BLD_DIR/libcudf/work/cpp/build
+  echo "Checking for build metrics log $LIBCUDF_BUILD_DIR/ninja_log.html"
+  if [[ -f "$LIBCUDF_BUILD_DIR/ninja_log.html" ]]; then
+      gpuci_logger "Copying build metrics results"
+      mkdir -p "$WORKSPACE/build-metrics"
+      cp "$LIBCUDF_BUILD_DIR/ninja_log.html" "$WORKSPACE/build-metrics/BuildMetrics.html"
+  fi
 
   gpuci_logger "Build conda pkg for libcudf_kafka"
   gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/libcudf_kafka $CONDA_BUILD_ARGS

diff --git a/ci/cpu/prebuild.sh b/ci/cpu/prebuild.sh
@@ -3,31 +3,11 @@
 # Copyright (c) 2020, NVIDIA CORPORATION.
 set -e
 
-DEFAULT_CUDA_VER="11.5"
-
-#Always upload cudf Python package
+#Always upload cudf packages
 export UPLOAD_CUDF=1
-
-#Upload libcudf once per CUDA
-if [[ "$PYTHON" == "3.7" ]]; then
-    export UPLOAD_LIBCUDF=1
-else
-    export UPLOAD_LIBCUDF=0
-fi
-
-# upload cudf_kafka for all versions of Python
-if [[ "$CUDA" == "${DEFAULT_CUDA_VER}" ]]; then
-    export UPLOAD_CUDF_KAFKA=1
-else
-    export UPLOAD_CUDF_KAFKA=0
-fi
-
-#We only want to upload libcudf_kafka once per python/CUDA combo
-if [[ "$PYTHON" == "3.7" ]] && [[ "$CUDA" == "${DEFAULT_CUDA_VER}" ]]; then
-    export UPLOAD_LIBCUDF_KAFKA=1
-else
-    export UPLOAD_LIBCUDF_KAFKA=0
-fi
+export UPLOAD_LIBCUDF=1
+export UPLOAD_CUDF_KAFKA=1
+export UPLOAD_LIBCUDF_KAFKA=1
 
 if [[ -z "$PROJECT_FLASH" || "$PROJECT_FLASH" == "0" ]]; then
     #If project flash is not activate, always build both

diff --git a/ci/cpu/upload.sh b/ci/cpu/upload.sh
@@ -12,7 +12,7 @@ export GPUCI_RETRY_SLEEP=30
 export LABEL_OPTION=${LABEL_OPTION:-"--label main"}
 
 # Skip uploads unless BUILD_MODE == "branch"
-if [ ${BUILD_MODE} != "branch" ]; then
+if [ "${BUILD_MODE}" != "branch" ]; then
   echo "Skipping upload"
   return 0
 fi

diff --git a/ci/gpu/build.sh b/ci/gpu/build.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
-# Copyright (c) 2018-2020, NVIDIA CORPORATION.
+# Copyright (c) 2018-2021, NVIDIA CORPORATION.
 ##############################################
 # cuDF GPU build and test script for CI      #
 ##############################################
@@ -31,7 +31,10 @@ export GIT_DESCRIBE_TAG=`git describe --tags`
 export MINOR_VERSION=`echo $GIT_DESCRIBE_TAG | grep -o -E '([0-9]+\.[0-9]+)'`
 
 # Dask & Distributed git tag
-export DASK_DISTRIBUTED_GIT_TAG='2021.11.2'
+export DASK_DISTRIBUTED_GIT_TAG='2022.01.0'
+
+# ucx-py version
+export UCX_PY_VERSION='0.24.*'
 
 ################################################################################
 # TRAP - Setup trap for removing jitify cache
@@ -83,10 +86,10 @@ gpuci_mamba_retry install -y \
                   "rapids-notebook-env=$MINOR_VERSION.*" \
                   "dask-cuda=${MINOR_VERSION}" \
                   "rmm=$MINOR_VERSION.*" \
-                  "ucx-py=0.23.*"
+                  "ucx-py=${UCX_PY_VERSION}"
 
 # https://docs.rapids.ai/maintainers/depmgmt/
-# gpuci_mamba_retry remove --force rapids-build-env rapids-notebook-env
+# gpuci_conda_retry remove --force rapids-build-env rapids-notebook-env
 # gpuci_mamba_retry install -y "your-pkg=1.0.0"
 
 
@@ -121,7 +124,7 @@ if [[ -z "$PROJECT_FLASH" || "$PROJECT_FLASH" == "0" ]]; then
     ################################################################################
 
     gpuci_logger "Build from source"
-    if [[ ${BUILD_MODE} == "pull-request" ]]; then
+    if [[ "${BUILD_MODE}" == "pull-request" ]]; then
         "$WORKSPACE/build.sh" clean libcudf cudf dask_cudf libcudf_kafka cudf_kafka benchmarks tests --ptds
     else
         "$WORKSPACE/build.sh" clean libcudf cudf dask_cudf libcudf_kafka cudf_kafka benchmarks tests -l --ptds
@@ -166,16 +169,46 @@ else
     gpuci_logger "Check GPU usage"
     nvidia-smi
 
-    gpuci_logger "GoogleTests"
     set -x
     cd $LIB_BUILD_DIR
 
+    gpuci_logger "GoogleTests"
+
     for gt in gtests/* ; do
         test_name=$(basename ${gt})
         echo "Running GoogleTest $test_name"
         ${gt} --gtest_output=xml:"$WORKSPACE/test-results/"
     done
 
+    # Copy libcudf build time results
+    echo "Checking for build time log $LIB_BUILD_DIR/ninja_log.xml"
+    if [[ -f "$LIB_BUILD_DIR/ninja_log.xml" ]]; then
+        gpuci_logger "Copying build time results"
+        cp "$LIB_BUILD_DIR/ninja_log.xml" "$WORKSPACE/test-results/buildtimes-junit.xml"
+    fi
+
+    ################################################################################
+    # MEMCHECK - Run compute-sanitizer on GoogleTest (only in nightly builds)
+    ################################################################################
+    if [[ "$BUILD_MODE" == "branch" && "$BUILD_TYPE" == "gpu" ]]; then
+        if [[ "$COMPUTE_SANITIZER_ENABLE" == "true" ]]; then
+            gpuci_logger "Memcheck on GoogleTests with rmm_mode=cuda"
+            export GTEST_CUDF_RMM_MODE=cuda
+            COMPUTE_SANITIZER_CMD="compute-sanitizer --tool memcheck"
+            mkdir -p "$WORKSPACE/test-results/"
+            for gt in gtests/*; do
+                test_name=$(basename ${gt})
+                if [[ "$test_name" == "ERROR_TEST" ]]; then
+                  continue
+                fi
+                echo "Running GoogleTest $test_name"
+                ${COMPUTE_SANITIZER_CMD} ${gt} | tee "$WORKSPACE/test-results/${test_name}.cs.log"
+            done
+            unset GTEST_CUDF_RMM_MODE
+            # test-results/*.cs.log are processed in gpuci
+        fi
+    fi
+
     CUDF_CONDA_FILE=`find ${CONDA_ARTIFACT_PATH} -name "libcudf-*.tar.bz2"`
     CUDF_CONDA_FILE=`basename "$CUDF_CONDA_FILE" .tar.bz2` #get filename without extension
     CUDF_CONDA_FILE=${CUDF_CONDA_FILE//-/=} #convert to conda install
@@ -184,12 +217,12 @@ else
     KAFKA_CONDA_FILE=${KAFKA_CONDA_FILE//-/=} #convert to conda install
 
     gpuci_logger "Installing $CUDF_CONDA_FILE & $KAFKA_CONDA_FILE"
-    conda install -c ${CONDA_ARTIFACT_PATH} "$CUDF_CONDA_FILE" "$KAFKA_CONDA_FILE"
+    gpuci_mamba_retry install -c ${CONDA_ARTIFACT_PATH} "$CUDF_CONDA_FILE" "$KAFKA_CONDA_FILE"
 
     install_dask
 
     gpuci_logger "Build python libs from source"
-    if [[ ${BUILD_MODE} == "pull-request" ]]; then
+    if [[ "${BUILD_MODE}" == "pull-request" ]]; then
         "$WORKSPACE/build.sh" cudf dask_cudf cudf_kafka --ptds
     else
         "$WORKSPACE/build.sh" cudf dask_cudf cudf_kafka -l --ptds

diff --git a/ci/gpu/java.sh b/ci/gpu/java.sh
@@ -30,6 +30,9 @@ export CONDA_ARTIFACT_PATH="$WORKSPACE/ci/artifacts/cudf/cpu/.conda-bld/"
 export GIT_DESCRIBE_TAG=`git describe --tags`
 export MINOR_VERSION=`echo $GIT_DESCRIBE_TAG | grep -o -E '([0-9]+\.[0-9]+)'`
 
+# ucx-py version
+export UCX_PY_VERSION='0.24.*'
+
 ################################################################################
 # TRAP - Setup trap for removing jitify cache
 ################################################################################
@@ -74,19 +77,23 @@ conda config --show-sources
 conda list --show-channel-urls
 
 gpuci_logger "Install dependencies"
-gpuci_conda_retry install -y \
+gpuci_mamba_retry install -y \
                   "cudatoolkit=$CUDA_REL" \
                   "rapids-build-env=$MINOR_VERSION.*" \
                   "rapids-notebook-env=$MINOR_VERSION.*" \
                   "dask-cuda=${MINOR_VERSION}" \
                   "rmm=$MINOR_VERSION.*" \
-                  "ucx-py=0.23.*" \
+                  "ucx-py=${UCX_PY_VERSION}" \
                   "openjdk=8.*" \
                   "maven"
+# "mamba install openjdk" adds an activation script to set JAVA_HOME but this is
+# not triggered on installation. Re-activating the conda environment will set
+# this environment variable so that CMake can find JNI.
+conda activate rapids
 
 # https://docs.rapids.ai/maintainers/depmgmt/
 # gpuci_conda_retry remove --force rapids-build-env rapids-notebook-env
-# gpuci_conda_retry install -y "your-pkg=1.0.0"
+# gpuci_mamba_retry install -y "your-pkg=1.0.0"
 
 
 gpuci_logger "Check compiler versions"
@@ -127,7 +134,7 @@ KAFKA_CONDA_FILE=`basename "$KAFKA_CONDA_FILE" .tar.bz2` #get filename without e
 KAFKA_CONDA_FILE=${KAFKA_CONDA_FILE//-/=} #convert to conda install
 
 gpuci_logger "Installing $CUDF_CONDA_FILE & $KAFKA_CONDA_FILE"
-conda install -c ${CONDA_ARTIFACT_PATH} "$CUDF_CONDA_FILE" "$KAFKA_CONDA_FILE"
+gpuci_mamba_retry install -c ${CONDA_ARTIFACT_PATH} "$CUDF_CONDA_FILE" "$KAFKA_CONDA_FILE"
 
 install_dask