Updated MLIO version and added buildspec

Author: Nikhil Raverkar

ci/buildspec.yml

@@ -0,0 +1,60 @@
+version: 0.2
+
+phases:
+  install:
+    runtime-versions:
+      python: 3.8
+      docker: 19
+  pre_build:
+    commands:
+    - echo Pre-build started on `date`
+    - echo Installing dependencies...
+    - curl -LO http://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh
+    - bash Miniconda3-latest-Linux-x86_64.sh -bfp /miniconda3
+    - export PATH=/miniconda3/bin:${PATH}
+    - conda install python=3.8
+    - conda update -y conda
+    - python3 -m pip install pip==20.1
+    - python3 -m pip install .[test]
+  build:
+    commands:
+    - echo Build started on `date`
+    - echo Docker login...
+    - docker login -u $dockerhub_username -p $dockerhub_password
+    - echo Building the Docker image...
+    - docker build -t xgboost-container-base:$FRAMEWORK_VERSION-cpu-py3 -f docker/$FRAMEWORK_VERSION/base/Dockerfile.cpu .
+    - python3 setup.py bdist_wheel --universal
+    - docker build -t preprod-xgboost-container:$FRAMEWORK_VERSION-cpu-py3 -f docker/$FRAMEWORK_VERSION/final/Dockerfile.cpu .
+    - printf "FROM preprod-xgboost-container:$FRAMEWORK_VERSION-cpu-py3\nADD . /app\nWORKDIR /app\nRUN python3 -m pip install .[test]" > Dockerfile.test
+    - docker build -t test-xgboost-container -f Dockerfile.test .
+    - echo Running tox...
+    - docker run --rm -t test-xgboost-container sh -c 'tox -e ALL'
+    - echo Running container tests...
+    - pytest test/integration/local --docker-base-name preprod-xgboost-container --tag $FRAMEWORK_VERSION-cpu-py3 --py-version 3 --framework-version $FRAMEWORK_VERSION
+    - docker tag preprod-xgboost-container:$FRAMEWORK_VERSION-cpu-py3 $SM_ALPHA.dkr.ecr.us-west-2.amazonaws.com/sagemaker-xgboost:$FRAMEWORK_VERSION-cpu-py3
+    - docker tag preprod-xgboost-container:$FRAMEWORK_VERSION-cpu-py3 $SM_ALPHA.dkr.ecr.us-west-2.amazonaws.com/sagemaker-xgboost:$FRAMEWORK_VERSION
+  post_build:
+    commands:
+    - echo Build completed on `date`
+    - |
+      case $CODEBUILD_WEBHOOK_EVENT in
+        PULL_REQUEST_MERGED)
+          echo Logging in to Amazon ECR...
+          $(aws ecr get-login --no-include-email --region $AWS_DEFAULT_REGION)
+          echo Pushing the Docker image...
+          docker push $SM_ALPHA.dkr.ecr.us-west-2.amazonaws.com/sagemaker-xgboost:$FRAMEWORK_VERSION-cpu-py3 | grep -v -E "[0-9]{12}.dkr.ecr.\S+.amazonaws.com"
+          docker push $SM_ALPHA.dkr.ecr.us-west-2.amazonaws.com/sagemaker-xgboost:$FRAMEWORK_VERSION | grep -v -E "[0-9]{12}.dkr.ecr.\S+.amazonaws.com"
+          ;;
+        PULL_REQUEST_CREATED | PULL_REQUEST_UPDATED | PULL_REQUEST_REOPENED)
+          echo Logging in to Amazon ECR...
+          $(aws ecr get-login --no-include-email --region $AWS_DEFAULT_REGION)
+          echo Pushing the Docker image...
+          # pushes test tag for manual verification, requires cleanup in ECR every once in a while though
+          TEST_TAG=$SM_ALPHA.dkr.ecr.us-west-2.amazonaws.com/sagemaker-xgboost:${FRAMEWORK_VERSION}-cpu-py3-test
+          docker tag preprod-xgboost-container:$FRAMEWORK_VERSION-cpu-py3 ${TEST_TAG}
+          docker push ${TEST_TAG} | grep -v -E "[0-9]{12}.dkr.ecr.\S+.amazonaws.com"
+          ;;
+        *)
+          echo Undefined behavior for webhook event type $CODEBUILD_WEBHOOK_EVENT
+          ;;
+      esac 

docker/1.0-1/base/Dockerfile.cpu

@@ -1,5 +1,8 @@
 FROM ubuntu:16.04
 
+ARG PYARROW_VERSION=0.16.0
+ARG MLIO_VERSION=0.6.0
+ARG PYTHON_VERSION=3.7.10
 # Install python and other runtime dependencies
 RUN apt-get update && \
     apt-get -y install \
@@ -25,11 +28,11 @@ RUN echo 'installing miniconda' && \
 
 ENV PATH=/miniconda3/bin:${PATH}
 
-RUN conda install -c conda-forge python=3.6.13 && \
+RUN conda install -c conda-forge python=${PYTHON_VERSION} && \
     conda update -y conda && \
-    conda install pip=20.1 && \
-    conda install -c conda-forge pyarrow=0.14.1 && \
-    conda install -c mlio -c conda-forge mlio-py=0.1
+    python3 -m pip install --upgrade pip && \
+    conda install -c conda-forge pyarrow=${PYARROW_VERSION} && \
+    conda install -c mlio -c conda-forge mlio-py=${MLIO_VERSION}
 
 # Python won’t try to write .pyc or .pyo files on the import of source modules
 # Force stdin, stdout and stderr to be totally unbuffered. Good for logging

requirements.txt

@@ -1,21 +1,27 @@
 Flask==1.1.1  # sagemaker-containers requires flask 1.1.1
 PyYAML==5.4.1
-boto3==1.10.14
-botocore==1.13.14
-gunicorn<20.0.0
-cryptography==3.4.6
+Pillow==9.0.0
+boto3==1.17.52
+botocore==1.20.52
+cryptography==35.0.0
+gunicorn==19.10.0
 matplotlib==3.3.2
 multi-model-server==1.1.1
 numpy==1.19.2
 pandas==1.1.3
+protobuf==3.20.1
 psutil==5.6.7  # sagemaker-containers requires psutil 5.6.7
 python-dateutil==2.8.0
-requests<2.21
+requests==2.25.1
 retrying==1.3.3
-sagemaker-containers>=2.8.3,<2.9
+sagemaker-containers==2.8.6.post2
 sagemaker-inference==1.2.0
 scikit-learn==0.23.2
-scipy==1.2.2
-smdebug==0.4.13
-urllib3==1.25.9
-wheel
+scipy==1.5.3
+smdebug==1.0.7
+urllib3==1.26.5
+wheel==0.35.1
+jinja2==3.0.3
+itsdangerous==2.0.1
+MarkupSafe==2.1.1
+Werkzeug==0.15.6

src/sagemaker_xgboost_container/data_utils.py

@@ -299,10 +299,11 @@ def _get_csv_dmatrix_pipe_mode(pipe_path, csv_weights):
     :return: xgb.DMatrix or None
     """
     try:
-        dataset = [mlio.SageMakerPipe(pipe_path)]
-        reader = mlio.CsvReader(dataset=dataset,
-                                batch_size=BATCH_SIZE,
-                                header_row_index=None)
+        pipes_path = pipe_path if isinstance(pipe_path, list) else [pipe_path]
+        dataset = [mlio.SageMakerPipe(path) for path in pipes_path]
+        reader_params = mlio.DataReaderParams(dataset=dataset, batch_size=BATCH_SIZE)
+        csv_params = mlio.CsvParams(header_row_index=None)
+        reader = mlio.CsvReader(reader_params, csv_params)
 
         # Check if data is present in reader
         if reader.peek_example() is not None:
@@ -449,17 +450,15 @@ def get_recordio_protobuf_dmatrix(path, is_pipe=False):
     """
     try:
         if is_pipe:
-            dataset = [mlio.SageMakerPipe(path)]
-            reader = mlio.RecordIOProtobufReader(dataset=dataset,
-                                                 batch_size=BATCH_SIZE)
+            pipes_path = path if isinstance(path, list) else [path]
+            dataset = [mlio.SageMakerPipe(pipe_path) for pipe_path in pipes_path]
         else:
             dataset = mlio.list_files(path)
-            reader = mlio.RecordIOProtobufReader(dataset=dataset,
-                                                 batch_size=BATCH_SIZE)
-
+        reader_params = mlio.DataReaderParams(dataset=dataset, batch_size=BATCH_SIZE)
+        reader = mlio.RecordIOProtobufReader(reader_params)
         if reader.peek_example() is not None:
             # recordio-protobuf tensor may be dense (use numpy) or sparse (use scipy)
-            if type(reader.peek_example()['values']) is mlio.core.DenseTensor:
+            if type(reader.peek_example()['values']) is mlio.DenseTensor:
                 to_matrix = as_numpy
                 vstack = np.vstack
             else:

src/sagemaker_xgboost_container/encoder.py

@@ -83,7 +83,8 @@ def recordio_protobuf_to_dmatrix(string_like):  # type: (bytes) -> xgb.DMatrix
     """
     buf = bytes(string_like)
     dataset = [mlio.InMemoryStore(buf)]
-    reader = mlio.RecordIOProtobufReader(dataset=dataset, batch_size=100)
+    reader_params = mlio.DataReaderParams(dataset=dataset, batch_size=100)
+    reader = mlio.RecordIOProtobufReader(reader_params)
 
     if type(reader.peek_example()['values']) is mlio.core.DenseTensor:
         to_matrix = as_numpy

tox.ini

@@ -1,5 +1,5 @@
 [tox]
-envlist = {py36}-xgboost{1.0},flake8
+envlist = {py37}-xgboost{1.0},flake8
 
 [flake8]
 max-line-length = 120
@@ -15,8 +15,9 @@ deps =
     -r{toxinidir}/requirements.txt
     -r{toxinidir}/test-requirements.txt
 conda_deps=
-    pyarrow=0.14.1
-    mlio-py=0.1
+    pyarrow=0.16.0
+    mlio-py=0.6.0
+    tbb==2020.2
 conda_channels=
     conda-forge
     mlio