Fork the neuron DLC dockerfile

.github/workflows/ci.yaml

@@ -25,4 +25,4 @@ jobs:
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@v3
-    - run: docker build --file e2e2/test/images/Dockerfile.neuronx-tests .
+    - run: docker build --file e2e2/test/images/neuron/Dockerfile .

e2e2/test/cases/neuron/manifests/single-node-test-neuronx.yaml

@@ -15,7 +15,7 @@ spec:
         image: {{.NeuronTestImage}}
         command:
         - /bin/bash
-        - ./pytorch_tests/singleNodeTest.sh
+        - ./tests/singleNodeTest.sh
         imagePullPolicy: Always
         resources:
           limits:

e2e2/test/cases/neuron/neuron_test.go

@@ -34,7 +34,8 @@ func TestMPIJobPytorchTraining(t *testing.T) {
 			if *neuronTestImage == "" {
 				t.Fatal(fmt.Errorf("neuronTestImage must be set to run neuron single node test, use https://github.com/aws/aws-k8s-tester/blob/main/e2e2/test/images/Dockerfile.neuronx-tests to build the image and -neuronTestImage to set the image url"))
 			}
-			renderedNeuronSingleNodeManifest, err := fwext.RenderManifests(neuronSingleNodeManifest, neuronSingleNodeManifestTplVars{
+			var err error
+			renderedNeuronSingleNodeManifest, err = fwext.RenderManifests(neuronSingleNodeManifest, neuronSingleNodeManifestTplVars{
 				NeuronTestImage: *neuronTestImage,
 			})
 			if err != nil {

e2e2/test/images/neuron/Dockerfile

@@ -0,0 +1,193 @@
+FROM public.ecr.aws/docker/library/ubuntu:20.04
+
+LABEL maintainer="Amazon AI"
+LABEL dlc_major_version="1"
+
+# Neuron SDK components version numbers
+ARG NEURONX_DISTRIBUTED_VERSION=0.7.0
+ARG NEURONX_CC_VERSION=2.13.72.0
+ARG NEURONX_FRAMEWORK_VERSION=2.1.2.2.1.0
+ARG NEURONX_COLLECTIVES_LIB_VERSION=2.20.22.0-c101c322e
+ARG NEURONX_RUNTIME_LIB_VERSION=2.20.22.0-1b3ca6425
+ARG NEURONX_TOOLS_VERSION=2.17.1.0
+
+ARG PYTHON=python3.10
+ARG PYTHON_VERSION=3.10.12
+ARG PIP=pip3
+ARG OMPI_VERSION=4.1.5
+
+# This arg required to stop docker build waiting for region configuration while installing tz data from ubuntu 20
+ARG DEBIAN_FRONTEND=noninteractive
+
+# 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
+ENV PYTHONDONTWRITEBYTECODE=1
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONIOENCODING=UTF-8
+ENV LANG=C.UTF-8
+ENV LC_ALL=C.UTF-8
+ENV LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:/opt/aws/neuron/lib"
+ENV LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:/opt/amazon/efa/lib"
+ENV LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:/opt/amazon/efa/lib64"
+ENV LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:/opt/amazon/openmpi/lib64"
+ENV LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:/usr/local/lib"
+ENV PATH /opt/aws/neuron/bin/:$PATH
+# ENV SAGEMAKER_TRAINING_MODULE=sagemaker_pytorch_container.training:main
+ENV DGLBACKEND=pytorch
+
+RUN apt-get update \
+ && apt-get upgrade -y \
+ && apt-get install -y --no-install-recommends \
+    build-essential \
+    ca-certificates \
+    cmake \
+    curl \
+    emacs \
+    git \
+    jq \
+    libopencv-dev \
+    software-properties-common \
+    wget \
+    unzip \
+    vim \
+    zlib1g-dev \
+    openssl \
+    libssl-dev \
+    libsqlite3-dev \
+    libgdbm-dev \
+    libc6-dev \
+    libbz2-dev \
+    libncurses-dev \
+    tk-dev \
+    libffi-dev \
+    libcap-dev \
+    gnupg2 \
+    gpg-agent \
+ && rm -rf /var/lib/apt/lists/* \
+ && apt-get clean
+
+RUN echo "deb https://apt.repos.neuron.amazonaws.com focal main" > /etc/apt/sources.list.d/neuron.list
+RUN wget -qO - https://apt.repos.neuron.amazonaws.com/GPG-PUB-KEY-AMAZON-AWS-NEURON.PUB | apt-key add -
+
+RUN apt-get update \
+ && apt-get install -y \
+    aws-neuronx-tools=$NEURONX_TOOLS_VERSION \
+    aws-neuronx-collectives=$NEURONX_COLLECTIVES_LIB_VERSION \
+    aws-neuronx-runtime-lib=$NEURONX_RUNTIME_LIB_VERSION \
+ && rm -rf /var/lib/apt/lists/* \
+ && rm -rf /tmp/tmp* \
+ && apt-get clean
+
+# Install Open MPI
+RUN mkdir -p /tmp/openmpi \
+ && cd /tmp/openmpi \
+ && wget --quiet https://download.open-mpi.org/release/open-mpi/v4.1/openmpi-${OMPI_VERSION}.tar.gz \
+ && tar zxf openmpi-${OMPI_VERSION}.tar.gz \
+ && cd openmpi-${OMPI_VERSION} \
+ && ./configure --enable-orterun-prefix-by-default \
+ && make -j $(nproc) all \
+ && make install \
+ && ldconfig \
+ && rm -rf /tmp/openmpi
+
+# install Python
+RUN wget -q https://www.python.org/ftp/python/$PYTHON_VERSION/Python-$PYTHON_VERSION.tgz \
+ && tar -xzf Python-$PYTHON_VERSION.tgz \
+ && cd Python-$PYTHON_VERSION \
+ && ./configure --enable-shared --prefix=/usr/local \
+ && make -j $(nproc) && make install \
+ && cd .. && rm -rf ../Python-$PYTHON_VERSION* \
+ && ln -s /usr/local/bin/pip3 /usr/bin/pip \
+ && ln -s /usr/local/bin/$PYTHON /usr/local/bin/python \
+ && ${PIP} --no-cache-dir install --upgrade \
+    pip \
+    setuptools
+
+WORKDIR /
+
+# The ENV variables declared below are changed in the previous section
+# Grouping these ENV variables in the first section causes
+# ompi_info to fail. This is only observed in CPU containers
+ENV PATH="$PATH:/home/.openmpi/bin"
+ENV LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/home/.openmpi/lib/"
+RUN ompi_info --parsable --all | grep mpi_built_with_cuda_support:value
+
+# Copy workaround script for incorrect hostname
+COPY e2e2/test/images/neuron/common/changehostname.c /
+COPY e2e2/test/images/neuron/common/start_with_right_hostname.sh /usr/local/bin/start_with_right_hostname.sh
+COPY e2e2/test/images/neuron/common/deep_learning_container.py /usr/local/bin/deep_learning_container.py
+COPY e2e2/test/images/neuron/tests ./tests
+
+RUN ${PIP} install --no-cache-dir -U \
+    "bokeh>=2.3,<3" \
+    "awscli<2" \
+    scipy \
+    click \
+    "cryptography" \
+    psutil==5.6.7 \
+    dataset \
+    transformers==4.36.2 \
+    Pillow
+
+RUN mkdir -p /etc/pki/tls/certs && cp /etc/ssl/certs/ca-certificates.crt /etc/pki/tls/certs/ca-bundle.crt
+RUN ${PIP} config set global.extra-index-url https://pip.repos.neuron.amazonaws.com \
+ && ${PIP} install --force-reinstall torch-neuronx==$NEURONX_FRAMEWORK_VERSION --extra-index-url https://pip.repos.neuron.amazonaws.com \
+ && ${PIP} install --force-reinstall neuronx-cc==$NEURONX_CC_VERSION --extra-index-url https://pip.repos.neuron.amazonaws.com \
+ && ${PIP} install --force-reinstall --no-deps neuronx_distributed==$NEURONX_DISTRIBUTED_VERSION --extra-index-url https://pip.repos.neuron.amazonaws.com
+
+# attrs, neuronx-cc required: >=19.2.0, sagemaker <24,>=23.1.0
+# protobuf neuronx-cc<4, sagemaker-training >=3.9.2,<=3.20.3
+# awscli 1.25.47 has requirement docutils<0.17,>=0.10
+# etcd for kubernetes installation
+# awscli 1.27.127 has requirement rsa<4.8,>=3.1.2, but you have rsa 4.9.
+# awscli 1.27.127 requires urllib3 < 1.27, python-etcd requires urllib3 >= 1.7, latest urllib3 release is 2.0.2
+RUN ${PIP} install --no-cache-dir -U \
+    "attrs<24,>=23.1.0" \
+    "protobuf>=3.18.3,<=3.20.3" \
+    "docutils>=0.10,<0.17" \
+    "rsa<4.8,>=3.1.2" \
+    "python-etcd" \
+    "urllib3>=1.26.0,<1.27"
+
+# EFA Installer does apt get. Make sure to run apt update before that
+RUN apt-get update
+RUN cd $HOME \
+ && curl -O https://efa-installer.amazonaws.com/aws-efa-installer-latest.tar.gz \
+ && wget https://efa-installer.amazonaws.com/aws-efa-installer.key && gpg --import aws-efa-installer.key \
+ && cat aws-efa-installer.key | gpg --fingerprint \
+ && wget https://efa-installer.amazonaws.com/aws-efa-installer-latest.tar.gz.sig && gpg --verify ./aws-efa-installer-latest.tar.gz.sig \
+ && tar -xf aws-efa-installer-latest.tar.gz \
+ && cd aws-efa-installer \
+ && ./efa_installer.sh -y -g -d --skip-kmod --skip-limit-conf --no-verify \
+ && cd $HOME
+
+
+# Clean up after apt update
+RUN rm -rf /var/lib/apt/lists/* \
+ && rm -rf /tmp/tmp* \
+ && apt-get clean
+
+# Install some common packages used by training scripts
+# torchvision needed for MLP. since it depends on torch and torch neuron/torch
+# is already installed install it with nodeps
+RUN pip3 install --no-cache-dir --no-deps -U \
+    torchvision==0.16.*
+
+RUN chmod +x /usr/local/bin/start_with_right_hostname.sh \
+ && chmod +x /usr/local/bin/deep_learning_container.py
+
+RUN HOME_DIR=/root \
+ && curl -o ${HOME_DIR}/oss_compliance.zip https://aws-dlinfra-utilities.s3.amazonaws.com/oss_compliance.zip \
+ && unzip ${HOME_DIR}/oss_compliance.zip -d ${HOME_DIR}/ \
+ && cp ${HOME_DIR}/oss_compliance/test/testOSSCompliance /usr/local/bin/testOSSCompliance \
+ && chmod +x /usr/local/bin/testOSSCompliance \
+ && chmod +x ${HOME_DIR}/oss_compliance/generate_oss_compliance.sh \
+ && ${HOME_DIR}/oss_compliance/generate_oss_compliance.sh ${HOME_DIR} ${PYTHON} \
+ && rm -rf ${HOME_DIR}/oss_compliance* \
+ && rm -rf /tmp/tmp*
+
+RUN curl -o /license.txt  https://aws-dlc-licenses.s3.amazonaws.com/pytorch-2.1/license.txt
+
+# Starts framework
+ENTRYPOINT ["bash", "-m", "start_with_right_hostname.sh"]
+CMD ["/bin/bash"]

e2e2/test/images/neuron/common/changehostname.c

@@ -0,0 +1,33 @@
+#include <stdio.h>
+#include <string.h>
+
+/**
+ * Copyright 2018-2023 Amazon.com, Inc. or its affiliates. All Rights Reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License"). You
+ * may not use this file except in compliance with the License. A copy of
+ * the License is located at
+ *
+ *     http://aws.amazon.com/apache2.0/
+ *
+ * or in the "license" file accompanying this file. This file is
+ * distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+ * ANY KIND, either express or implied. See the License for the specific
+ * language governing permissions and limitations under the License.
+ */
+
+/**
+ * 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;
+}