infra: Rename buildspec files. (#211)

buildspec-container-pr.yml

@@ -0,0 +1,13 @@
+version: 0.2
+
+phases:
+  pre_build:
+    commands:
+      - PR_NUM=$(echo $CODEBUILD_SOURCE_VERSION | grep -o '[0-9]\+')
+      - echo 'Pull request number:' $PR_NUM '. No value means this build is not from pull request.'
+
+  build:
+    commands:
+
+      - error_cmd="echo 'In order to make changes to the docker files, please, use https://github.com/aws/deep-learning-containers repository.' && exit 1"
+      - execute-command-if-has-matching-changes "$error_cmd" "docker/"

buildspec.yml

@@ -1,13 +1,94 @@
 version: 0.2
 
+env:
+  variables:
+    FRAMEWORK_VERSION: '1.4.0'
+    CPU_INSTANCE_TYPE: 'ml.c4.xlarge'
+    GPU_INSTANCE_TYPE: 'ml.p2.8xlarge'
+    ECR_REPO: 'sagemaker-test'
+    GITHUB_REPO: 'sagemaker-pytorch-container'
+    DLC_ACCOUNT: '763104351884'
+    SETUP_FILE: 'setup_cmds.sh'
+    SETUP_CMDS: '#!/bin/bash\npip install --upgrade pip\npip install -U -e .\npip install -U -e .[test]'
+
 phases:
   pre_build:
     commands:
+      - start-dockerd
+      - ACCOUNT=$(aws --region $AWS_DEFAULT_REGION sts --endpoint-url https://sts.$AWS_DEFAULT_REGION.amazonaws.com get-caller-identity --query 'Account' --output text)
+      - PREPROD_IMAGE="$ACCOUNT.dkr.ecr.$AWS_DEFAULT_REGION.amazonaws.com/$ECR_REPO"
       - PR_NUM=$(echo $CODEBUILD_SOURCE_VERSION | grep -o '[0-9]\+')
+      - BUILD_ID="$(echo $CODEBUILD_BUILD_ID | sed -e 's/:/-/g')"
       - echo 'Pull request number:' $PR_NUM '. No value means this build is not from pull request.'
 
   build:
     commands:
+      - TOX_PARALLEL_NO_SPINNER=1
+      - PY_COLORS=0
+
+      # install
+      - pip3 install -U -e .[test]
+
+      # run linters
+      - tox -e flake8,twine
+
+      # run unit tests
+      - tox -e py27,py36,py37 test/unit
+
+      # define tags
+      - GENERIC_TAG="$FRAMEWORK_VERSION-pytorch-$BUILD_ID"
+      - DLC_CPU_TAG="$FRAMEWORK_VERSION-dlc-cpu-$BUILD_ID"
+      - DLC_GPU_TAG="$FRAMEWORK_VERSION-dlc-gpu-$BUILD_ID"
+
+      # run local CPU integration tests (build and push the image to ECR repo)
+      - test_cmd="pytest test/integration/local --build-image --push-image --dockerfile-type pytorch --region $AWS_DEFAULT_REGION --docker-base-name $ECR_REPO --aws-id $ACCOUNT --framework-version $FRAMEWORK_VERSION --processor cpu --tag $GENERIC_TAG"
+      - execute-command-if-has-matching-changes "$test_cmd" "test/" "src/*.py" "setup.py" "setup.cfg" "buildspec.yml" "lib/*"
+      - test_cmd="pytest test/integration/local --build-image --push-image --dockerfile-type dlc.cpu --region $AWS_DEFAULT_REGION --docker-base-name $ECR_REPO --aws-id $ACCOUNT --framework-version $FRAMEWORK_VERSION --processor cpu --tag $DLC_CPU_TAG"
+      - execute-command-if-has-matching-changes "$test_cmd" "test/" "src/*.py" "setup.py" "setup.cfg" "buildspec.yml" "lib/*"
+
+      # launch remote GPU instance
+      - prefix='ml.'
+      - instance_type=${GPU_INSTANCE_TYPE#"$prefix"}
+      - create-key-pair
+      - launch-ec2-instance --instance-type $instance_type --ami-name dlami-ubuntu-latest
+
+      # build DLC GPU image because the base DLC image is too big and takes too long to build as part of the test
+      - python3 setup.py sdist
+      - build_dir="test/container/$FRAMEWORK_VERSION"
+      - $(aws ecr get-login --registry-ids $DLC_ACCOUNT --no-include-email --region $AWS_DEFAULT_REGION)
+      - docker build -f "$build_dir/Dockerfile.dlc.gpu" -t $PREPROD_IMAGE:$DLC_GPU_TAG --build-arg region=$AWS_DEFAULT_REGION .
+      # push DLC GPU image to ECR
+      - $(aws ecr get-login --registry-ids $ACCOUNT --no-include-email --region $AWS_DEFAULT_REGION)
+      - docker push $PREPROD_IMAGE:$DLC_GPU_TAG
+
+      # run GPU local integration tests
+      - printf "$SETUP_CMDS" > $SETUP_FILE
+      # no reason to rebuild the image again since it was already built and pushed to ECR during CPU tests
+      - generic_cmd="pytest test/integration/local --region $AWS_DEFAULT_REGION --docker-base-name $ECR_REPO --aws-id $ACCOUNT --framework-version $FRAMEWORK_VERSION --processor gpu --tag $GENERIC_TAG"
+      - test_cmd="remote-test --github-repo $GITHUB_REPO --test-cmd \"$generic_cmd\" --setup-file $SETUP_FILE --pr-number \"$PR_NUM\""
+      - execute-command-if-has-matching-changes "$test_cmd" "test/" "src/*.py" "setup.py" "setup.cfg" "buildspec.yml" "lib/*"
+      - dlc_cmd="pytest test/integration/local --region $AWS_DEFAULT_REGION --docker-base-name $ECR_REPO --aws-id $ACCOUNT --framework-version $FRAMEWORK_VERSION --processor gpu --tag $DLC_GPU_TAG"
+      - test_cmd="remote-test --github-repo $GITHUB_REPO --test-cmd \"$dlc_cmd\" --setup-file $SETUP_FILE --pr-number \"$PR_NUM\" --skip-setup"
+      - execute-command-if-has-matching-changes "$test_cmd" "test/" "src/*.py" "setup.py" "setup.cfg" "buildspec.yml" "lib/*"
+
+      # run CPU sagemaker integration tests
+      - test_cmd="pytest -n 10 test/integration/sagemaker --region $AWS_DEFAULT_REGION --docker-base-name $ECR_REPO --aws-id $ACCOUNT --framework-version $FRAMEWORK_VERSION --processor cpu --instance-type $CPU_INSTANCE_TYPE --tag $GENERIC_TAG"
+      - execute-command-if-has-matching-changes "$test_cmd" "test/" "src/*.py" "setup.py" "setup.cfg" "buildspec.yml" "lib/*"
+      - test_cmd="pytest -n 10 test/integration/sagemaker --region $AWS_DEFAULT_REGION --docker-base-name $ECR_REPO --aws-id $ACCOUNT --framework-version $FRAMEWORK_VERSION --processor cpu --instance-type $CPU_INSTANCE_TYPE --tag $DLC_CPU_TAG"
+      - execute-command-if-has-matching-changes "$test_cmd" "test/" "src/*.py" "setup.py" "setup.cfg" "buildspec.yml" "lib/*"
+
+      # run GPU sagemaker integration tests
+      - test_cmd="pytest -n 10 test/integration/sagemaker --region $AWS_DEFAULT_REGION --docker-base-name $ECR_REPO --aws-id $ACCOUNT --framework-version $FRAMEWORK_VERSION --processor gpu --instance-type $GPU_INSTANCE_TYPE --tag $GENERIC_TAG"
+      - execute-command-if-has-matching-changes "$test_cmd" "test/" "src/*.py" "setup.py" "setup.cfg" "buildspec.yml" "lib/*"
+      - test_cmd="pytest -n 10 test/integration/sagemaker --region $AWS_DEFAULT_REGION --docker-base-name $ECR_REPO --aws-id $ACCOUNT --framework-version $FRAMEWORK_VERSION --processor gpu --instance-type $GPU_INSTANCE_TYPE --tag $DLC_GPU_TAG"
+      - execute-command-if-has-matching-changes "$test_cmd" "test/" "src/*.py" "setup.py" "setup.cfg" "buildspec.yml" "lib/*"
+
+    finally:
+      # shut down remote GPU instance
+      - cleanup-gpu-instances
+      - cleanup-key-pairs
 
-      - error_cmd="echo 'In order to make changes to the docker files, please, use https://github.com/aws/deep-learning-containers repository.' && exit 1"
-      - execute-command-if-has-matching-changes "$error_cmd" "docker/"
+      # remove ECR image
+      - aws ecr batch-delete-image --repository-name $ECR_REPO --region $AWS_DEFAULT_REGION --image-ids imageTag=$GENERIC_TAG
+      - aws ecr batch-delete-image --repository-name $ECR_REPO --region $AWS_DEFAULT_REGION --image-ids imageTag=$DLC_CPU_TAG
+      - aws ecr batch-delete-image --repository-name $ECR_REPO --region $AWS_DEFAULT_REGION --image-ids imageTag=$DLC_GPU_TAG

lib/changehostname.c

@@ -0,0 +1,18 @@
+#include <stdio.h>
+#include <string.h>
+
+/*
+ * Modifies gethostname to return algo-1, algo-2, etc. when running on SageMaker.
+ *
+ * Without this gethostname() on SageMaker returns 'aws', leading NCCL/MPI to think there is only one host,
+ * not realizing that it needs to use NET/Socket.
+ *
+ * When docker container starts we read 'current_host' value  from /opt/ml/input/config/resourceconfig.json
+ * and replace PLACEHOLDER_HOSTNAME with it before compiling this code into a shared library.
+ */
+int gethostname(char *name, size_t len)
+{
+  const char *val = PLACEHOLDER_HOSTNAME;
+  strncpy(name, val, len);
+  return 0;
+}

lib/start_with_right_hostname.sh

@@ -0,0 +1,11 @@
+#!/usr/bin/env bash
+
+if [[ "$1" = "train" ]]; then
+     CURRENT_HOST=$(jq .current_host  /opt/ml/input/config/resourceconfig.json)
+     sed -ie "s/PLACEHOLDER_HOSTNAME/$CURRENT_HOST/g" changehostname.c
+     gcc -o changehostname.o -c -fPIC -Wall changehostname.c
+     gcc -o libchangehostname.so -shared -export-dynamic changehostname.o -ldl
+     LD_PRELOAD=/libchangehostname.so train
+else
+     eval "$@"
+fi

test/container/1.4.0/Dockerfile.dlc.cpu

@@ -1,8 +1,8 @@
 ARG region
 from 763104351884.dkr.ecr.$region.amazonaws.com/pytorch-training:1.4.0-cpu-py2
 
-COPY docker/build_artifacts/changehostname.c /
-COPY docker/build_artifacts/start_with_right_hostname.sh /usr/local/bin/start_with_right_hostname.sh
+COPY lib/changehostname.c /
+COPY lib/start_with_right_hostname.sh /usr/local/bin/start_with_right_hostname.sh
 RUN chmod +x /usr/local/bin/start_with_right_hostname.sh
 
 COPY dist/sagemaker_pytorch_training-*.tar.gz /sagemaker_pytorch_training.tar.gz

test/container/1.4.0/Dockerfile.dlc.gpu

@@ -1,8 +1,8 @@
 ARG region
 from 763104351884.dkr.ecr.$region.amazonaws.com/pytorch-training:1.4.0-gpu-py3
 
-COPY docker/build_artifacts/changehostname.c /
-COPY docker/build_artifacts/start_with_right_hostname.sh /usr/local/bin/start_with_right_hostname.sh
+COPY lib/changehostname.c /
+COPY lib/start_with_right_hostname.sh /usr/local/bin/start_with_right_hostname.sh
 RUN chmod +x /usr/local/bin/start_with_right_hostname.sh
 
 COPY dist/sagemaker_pytorch_training-*.tar.gz /sagemaker_pytorch_training.tar.gz

test/container/1.4.0/Dockerfile.pytorch

@@ -4,8 +4,8 @@ RUN apt-get update \
  && apt-get install -y --no-install-recommends jq \
  && rm -rf /var/lib/apt/lists/*
 
-COPY docker/build_artifacts/changehostname.c /
-COPY docker/build_artifacts/start_with_right_hostname.sh /usr/local/bin/start_with_right_hostname.sh
+COPY lib/changehostname.c /
+COPY lib/start_with_right_hostname.sh /usr/local/bin/start_with_right_hostname.sh
 RUN chmod +x /usr/local/bin/start_with_right_hostname.sh
 
 COPY dist/sagemaker_pytorch_training-*.tar.gz /sagemaker_pytorch_training.tar.gz