Example: Creating a Custom Image for Training (MindSpore and GPUs)

Scenario

In this example, write a Dockerfile to create a custom image on a Linux x86_64 server running Ubuntu 18.04.

Create a container image with the following configurations and use the image to create a GPU-powered training job on ModelArts:

• ubuntu-18.04
• cuda-11.1
• python-3.7.13
• mlnx ofed-5.4
• mindspore gpu-1.8.1

Procedure

Before using a custom image to create a training job, you need to be familiar with Docker and have development experience.

• Prerequisites
• Step 1 Creating an OBS Bucket and Folder
• Step 2 Creating a Dataset and Uploading It to OBS
• Step 3 Preparing the Training Script and Uploading It to OBS
• Step 4 Preparing a Server
• Step 5 Creating a Custom Image
• Step 6 Uploading the Image to SWR
• Step 7 Creating a Training Job on ModelArts

Prerequisites

You have registered a Huawei Cloud account. The account is not in arrears or frozen.

Step 1 Creating an OBS Bucket and Folder

Create a bucket and folders in OBS for storing the sample dataset and training code. Table 1 lists the folders to be created. Replace the bucket name and folder names in the example with actual names.

For details, see Creating a Bucket and Creating a Folder.

Ensure that the OBS and ModelArts are in the same region.

Step 2 Creating a Dataset and Uploading It to OBS

Go to http://www.cs.toronto.edu/~kriz/cifar.html, download CIFAR-10 binary version (suitable for C programs), decompress it, and upload the decompressed data to obs://test-modelarts/mindspore-gpu/cifar-10-batches-bin/ in the OBS bucket, which is as follows.

Figure 1 Datasets

Step 3 Preparing the Training Script and Uploading It to OBS

Obtain the ResNet file and script run_mpi.sh and upload them to obs://test-modelarts/mindspore-gpu/resnet/ in the OBS bucket.

Download the ResNet file from https://gitee.com/mindspore/models/tree/r1.8/official/cv/resnet.

run_mpi.sh is as follows:

#!/bin/bash
MY_HOME=/home/ma-user

MY_SSHD_PORT=${MY_SSHD_PORT:-"36666"}

MY_MPI_BTL_TCP_IF=${MY_MPI_BTL_TCP_IF:-"eth0,bond0"}

MY_TASK_INDEX=${MA_TASK_INDEX:-${VC_TASK_INDEX:-${VK_TASK_INDEX}}}

MY_MPI_SLOTS=${MY_MPI_SLOTS:-"${MA_NUM_GPUS}"}

MY_MPI_TUNE_FILE="${MY_HOME}/env_for_user_process"

if [ -z ${MY_MPI_SLOTS} ]; then
    echo "[run_mpi] MY_MPI_SLOTS is empty, set it be 1"
    MY_MPI_SLOTS="1"
fi

printf "MY_HOME: ${MY_HOME}\nMY_SSHD_PORT: ${MY_SSHD_PORT}\nMY_MPI_BTL_TCP_IF: ${MY_MPI_BTL_TCP_IF}\nMY_TASK_INDEX: ${MY_TASK_INDEX}\nMY_MPI_SLOTS: ${MY_MPI_SLOTS}\n"

env | grep -E '^MA_|^SHARED_|^S3_|^PATH|^VC_WORKER_|^SCC|^CRED' | grep -v '=$' > ${MY_MPI_TUNE_FILE}
# add -x to each line
sed -i 's/^/-x /' ${MY_MPI_TUNE_FILE}

sed -i "s|{{MY_SSHD_PORT}}|${MY_SSHD_PORT}|g" ${MY_HOME}/etc/ssh/sshd_config

# start sshd service
bash -c "$(which sshd) -f ${MY_HOME}/etc/ssh/sshd_config"

# confirm the sshd is up
netstat -anp | grep LIS | grep ${MY_SSHD_PORT}

if [ $MY_TASK_INDEX -eq 0 ]; then
    # generate the hostfile of mpi
    for ((i=0; i<$MA_NUM_HOSTS; i++))
    do
        eval hostname=${MA_VJ_NAME}-${MA_TASK_NAME}-${i}.${MA_VJ_NAME}
        echo "[run_mpi] hostname: ${hostname}"

        ip=""
        while [ -z "$ip" ]; do
            ip=$(ping -c 1 ${hostname} | grep "PING" | sed -E 's/PING .* .([0-9.]+). .*/\1/g')
            sleep 1
        done
        echo "[run_mpi] resolved ip: ${ip}"

        # test the sshd is up
        while :
        do
            if [ cat < /dev/null >/dev/tcp/${ip}/${MY_SSHD_PORT} ]; then
                break
            fi
            sleep 1
        done

        echo "[run_mpi] the sshd of ip ${ip} is up"

        echo "${ip} slots=$MY_MPI_SLOTS" >> ${MY_HOME}/hostfile
    done

    printf "[run_mpi] hostfile:\n`cat ${MY_HOME}/hostfile`\n"
fi

RET_CODE=0

if [ $MY_TASK_INDEX -eq 0 ]; then

    echo "[run_mpi] start exec command time: "$(date +"%Y-%m-%d-%H:%M:%S")

    np=$(( ${MA_NUM_HOSTS} * ${MY_MPI_SLOTS} ))

    echo "[run_mpi] command: mpirun -np ${np} -hostfile ${MY_HOME}/hostfile -mca plm_rsh_args \"-p ${MY_SSHD_PORT}\" -tune ${MY_MPI_TUNE_FILE} ... $@"

    # execute mpirun at worker-0
    # mpirun
    mpirun \
        -np ${np} \
        -hostfile ${MY_HOME}/hostfile \
        -mca plm_rsh_args "-p ${MY_SSHD_PORT}" \
        -tune ${MY_MPI_TUNE_FILE} \
        -bind-to none -map-by slot \
        -x NCCL_DEBUG=INFO -x NCCL_SOCKET_IFNAME=${MY_MPI_BTL_TCP_IF} -x NCCL_SOCKET_FAMILY=AF_INET \
        -x HOROVOD_MPI_THREADS_DISABLE=1 \
        -x LD_LIBRARY_PATH \
        -mca pml ob1 -mca btl ^openib -mca plm_rsh_no_tree_spawn true \
        "$@"

    RET_CODE=$?

    if [ $RET_CODE -ne 0 ]; then
        echo "[run_mpi] exec command failed, exited with $RET_CODE"
    else
        echo "[run_mpi] exec command successfully, exited with $RET_CODE"
    fi

    # stop 1...N worker by killing the sleep proc
    sed -i '1d' ${MY_HOME}/hostfile
    if [ `cat ${MY_HOME}/hostfile | wc -l` -ne 0 ]; then
        echo "[run_mpi] stop 1 to (N - 1) worker by killing the sleep proc"

        sed -i 's/${MY_MPI_SLOTS}/1/g' ${MY_HOME}/hostfile
        printf "[run_mpi] hostfile:\n`cat ${MY_HOME}/hostfile`\n"

        mpirun \
        --hostfile ${MY_HOME}/hostfile \
        --mca btl_tcp_if_include ${MY_MPI_BTL_TCP_IF} \
        --mca plm_rsh_args "-p ${MY_SSHD_PORT}" \
        -x PATH -x LD_LIBRARY_PATH \
        pkill sleep \
        > /dev/null 2>&1
    fi

    echo "[run_mpi] exit time: "$(date +"%Y-%m-%d-%H:%M:%S")
else
    echo "[run_mpi] the training log is in worker-0"
    sleep 365d
    echo "[run_mpi] exit time: "$(date +"%Y-%m-%d-%H:%M:%S")
fi

exit $RET_CODE

The following figure shows obs://test-modelarts/mindspore-gpu/resnet/, including the ResNet file and run_mpi.sh.

Figure 2 ResNet file and run_mpi.sh

Step 4 Preparing a Server

Obtain a Linux x86_64 server running Ubuntu 18.04. Either an ECS or your local PC will do.

For details about how to purchase an ECS, see Purchasing and Logging In to a Linux ECS. Select a public image. An Ubuntu 18.04 image is recommended.

Figure 3 Creating an ECS using a public image (x86)

Step 5 Creating a Custom Image

Create a container image with the following configurations and use the image to create a training job on ModelArts:

• ubuntu-18.04
• cuda-11.1
• python-3.7.13
• mlnx ofed-5.4
• mindspore gpu-1.8.1

This section describes how to write a Dockerfile to create a custom image.

1. Install Docker.

   The following uses Linux x86_64 as an example to describe how to obtain a Docker installation package. For more details about how to install Docker, see official Docker documents. Run the following command to install Docker:

   curl -fsSL get.docker.com -o get-docker.sh
   sh get-docker.sh

   If the docker images command can be executed, Docker has been installed. In this case, skip this step.

2. Check the Docker Engine version. Run the following command:

   docker version | grep -A 1 Engine

   The following information is displayed:

    Engine:
     Version:          18.09.0

   

   Use the Docker engine of the preceding version or later to create a custom image.

3. Create a folder named context.

   mkdir -p context

4. Obtain the pip.conf file. In this example, the pip source provided by Huawei Mirrors is used, which is as follows:

   [global]
   index-url = https://repo.huaweicloud.com/repository/pypi/simple
   trusted-host = repo.huaweicloud.com
   timeout = 120

   

   To obtain pip.conf, switch to Huawei Mirrors https://mirrors.huaweicloud.com/home and search for pypi.

5. Download mindspore_gpu-1.8.1-cp37-cp37m-linux_x86_64.whl from https://ms-release.obs.cn-north-4.myhuaweicloud.com/1.8.1/MindSpore/gpu/x86_64/cuda-11.1/mindspore_gpu-1.8.1-cp37-cp37m-linux_x86_64.whl.
6. Download the Miniconda3 installation file.

   Download the Miniconda3 py37 4.12.0 installation file (Python 3.7.13) from https://repo.anaconda.com/miniconda/Miniconda3-py37_4.12.0-Linux-x86_64.sh.

7. Write the container image Dockerfile.
   Create an empty file named Dockerfile in the context folder and copy the following content to the file:

   # The server on which the container image is created must access the Internet.
   
   # Base container image at https://github.com/NVIDIA/nvidia-docker/wiki/CUDA
   #
   # https://docs.docker.com/develop/develop-images/multistage-build/#use-multi-stage-builds
   # require Docker Engine >= 17.05
   #
   # builder stage
   FROM nvidia/cuda:11.1.1-devel-ubuntu18.04 AS builder
   
   # The default user of the base container image is root.
   # USER root
   
   # Use the PyPI configuration obtained from Huawei Mirrors.
   RUN mkdir -p /root/.pip/
   COPY pip.conf /root/.pip/pip.conf
   
   # Copy the installation files to the /tmp directory in the base container image.
   COPY Miniconda3-py37_4.12.0-Linux-x86_64.sh /tmp
   COPY mindspore_gpu-1.8.1-cp37-cp37m-linux_x86_64.whl /tmp
   
   # https://conda.io/projects/conda/en/latest/user-guide/install/linux.html#installing-on-linux
   # Install Miniconda3 in the /home/ma-user/miniconda3 directory of the base container image.
   RUN bash /tmp/Miniconda3-py37_4.12.0-Linux-x86_64.sh -b -p /home/ma-user/miniconda3
   
   # Install the whl file using default Miniconda3 Python environment /home/ma-user/miniconda3/bin/pip.
   RUN cd /tmp && \
       /home/ma-user/miniconda3/bin/pip install --no-cache-dir \
       /tmp/mindspore_gpu-1.8.1-cp37-cp37m-linux_x86_64.whl \
       easydict PyYAML
   
   # Create the container image.
   FROM nvidia/cuda:11.1.1-cudnn8-runtime-ubuntu18.04
   
   COPY MLNX_OFED_LINUX-5.4-3.5.8.0-ubuntu18.04-x86_64.tgz /tmp
   
   # Install the vim, cURL, net-tools, MLNX_OFED, and SSH tools obtained from Huawei Mirrors.
   RUN cp -a /etc/apt/sources.list /etc/apt/sources.list.bak && \
       sed -i "s@http://.*archive.ubuntu.com@http://repo.huaweicloud.com@g" /etc/apt/sources.list && \
       sed -i "s@http://.*security.ubuntu.com@http://repo.huaweicloud.com@g" /etc/apt/sources.list && \
       echo > /etc/apt/apt.conf.d/00skip-verify-peer.conf "Acquire { https::Verify-Peer false }" && \
       apt-get update && \
       apt-get install -y vim curl net-tools iputils-ping libfile-find-rule-perl-perl \
       openssh-client openssh-server && \
       ssh -V && \
       mkdir -p /run/sshd && \
       # mlnx ofed
       apt-get install -y python libfuse2 dpatch libnl-3-dev autoconf libnl-route-3-dev pciutils libnuma1 libpci3 m4 libelf1 debhelper automake graphviz bison lsof kmod libusb-1.0-0 swig libmnl0 autotools-dev flex chrpath libltdl-dev && \
       cd /tmp && \
       tar -xvf MLNX_OFED_LINUX-5.4-3.5.8.0-ubuntu18.04-x86_64.tgz && \
       MLNX_OFED_LINUX-5.4-3.5.8.0-ubuntu18.04-x86_64/mlnxofedinstall --user-space-only --basic --without-fw-update -q && \
       cd - && \
       rm -rf /tmp/* && \
       apt-get clean && \
       mv /etc/apt/sources.list.bak /etc/apt/sources.list && \
       rm /etc/apt/apt.conf.d/00skip-verify-peer.conf
   
   # Install the Open MPI 3.0.0 file obtained from Horovod v0.22.1.
   # https://github.com/horovod/horovod/blob/v0.22.1/docker/horovod/Dockerfile
   # https://github.com/horovod/horovod/files/1596799/openmpi-3.0.0-bin.tar.gz
   COPY openmpi-3.0.0-bin.tar.gz /tmp
   RUN cd /usr/local && \
       tar -zxf /tmp/openmpi-3.0.0-bin.tar.gz && \
       ldconfig && \
       mpirun --version
   
   # Add user ma-user (UID = 1000, GID = 100).
   # A user group whose GID is 100 exists in the basic container image. User ma-user can directly run the following command:
   RUN useradd -m -d /home/ma-user -s /bin/bash -g 100 -u 1000 ma-user
   
   # Copy the /home/ma-user/miniconda3 directory from the builder stage to the directory with the same name in the current container image.
   COPY --chown=ma-user:100 --from=builder /home/ma-user/miniconda3 /home/ma-user/miniconda3
   
   # Configure the default user and working directory of the container image.
   USER ma-user
   WORKDIR /home/ma-user
   
   # Configure sshd to support SSH password-free login.
   RUN MA_HOME=/home/ma-user && \
       # setup sshd dir
       mkdir -p ${MA_HOME}/etc && \
       ssh-keygen -f ${MA_HOME}/etc/ssh_host_rsa_key -N '' -t rsa  && \
       mkdir -p ${MA_HOME}/etc/ssh ${MA_HOME}/var/run  && \
       # setup sshd config (listen at {{MY_SSHD_PORT}} port)
       echo "Port {{MY_SSHD_PORT}}\n\
   HostKey ${MA_HOME}/etc/ssh_host_rsa_key\n\
   AuthorizedKeysFile ${MA_HOME}/.ssh/authorized_keys\n\
   PidFile ${MA_HOME}/var/run/sshd.pid\n\
   StrictModes no\n\
   UsePAM no" > ${MA_HOME}/etc/ssh/sshd_config && \
       # generate ssh key
       ssh-keygen -t rsa -f ${MA_HOME}/.ssh/id_rsa -P '' && \
       cat ${MA_HOME}/.ssh/id_rsa.pub >> ${MA_HOME}/.ssh/authorized_keys && \
       # disable ssh host key checking for all hosts
       echo "Host *\n\
     StrictHostKeyChecking no" > ${MA_HOME}/.ssh/config
   
   # Configure the preset environment variables of the container image.
   # Set PYTHONUNBUFFERED to 1 to prevent log loss.
   ENV PATH=/home/ma-user/miniconda3/bin:$PATH \
       LD_LIBRARY_PATH=/usr/local/cuda/lib64:/usr/lib/x86_64-linux-gnu:$LD_LIBRARY_PATH \
       PYTHONUNBUFFERED=1

   For details about how to write a Dockerfile, see official Docker documents.

8. Download MLNX_OFED_LINUX-5.4-3.5.8.0-ubuntu18.04-x86_64.tgz.

   Go to https://network.nvidia.com/products/infiniband-drivers/linux/mlnx_ofed/, click Download, set Version to 5.4-3.5.8.0-LTS, OSDistributionVersion to Ubuntu 18.04, and Architecture to x86_64, and download MLNX_OFED_LINUX-5.4-3.5.8.0-ubuntu18.04-x86_64.tgz.

9. Download openmpi-3.0.0-bin.tar.gz.

   Download openmpi-3.0.0-bin.tar.gz from https://github.com/horovod/horovod/files/1596799/openmpi-3.0.0-bin.tar.gz.

10. Store the Dockerfile and Miniconda3 installation file in the context folder, which is as follows:

    context
    ├── Dockerfile
    ├── MLNX_OFED_LINUX-5.4-3.5.8.0-ubuntu18.04-x86_64.tgz
    ├── Miniconda3-py37_4.12.0-Linux-x86_64.sh
    ├── mindspore_gpu-1.8.1-cp37-cp37m-linux_x86_64.whl
    ├── openmpi-3.0.0-bin.tar.gz
    └── pip.conf

11. Create the container image. Run the following command in the directory where the Dockerfile is stored to build the container image mindspore:1.8.1-ofed-cuda11.1:

    1	docker build . -t mindspore:1.8.1-ofed-cuda11.1

    The following log shows that the image has been created.

    Successfully tagged mindspore:1.8.1-ofed-cuda11.1

Step 6 Uploading the Image to SWR

1. Log in to the SWR console and select the target region.
   Figure 4 SWR console
   
2. Click Create Organization in the upper right corner and enter an organization name to create an organization. Customize the organization name. Replace the organization name deep-learning in subsequent commands with the actual organization name.
   Figure 5 Creating an organization
   
3. Click Generate Login Command in the upper right corner to obtain a login command.
   Figure 6 Login Command
   
4. Log in to the local environment as the root user and enter the login command.
5. Upload the image to SWR.
   1. Tag the uploaded image.

      # Replace the region, domain, as well as organization name deep-learning with the actual values.
      sudo docker tag mindspore:1.8.1-ofed-cuda11.1 swr.{region-id}.{domain}/deep-learning/mindspore:1.8.1-ofed-cuda11.1

   2. Run the following command to upload the image:

      # Replace the region, domain, as well as organization name deep-learning with the actual values.
      sudo docker push swr.{region-id}.{domain}/deep-learning/mindspore:1.8.1-ofed-cuda11.1

6. After the image is uploaded, choose My Images in navigation pane on the left of the SWR console to view the uploaded custom images.

Step 7 Creating a Training Job on ModelArts

1. Log in to the ModelArts management console, check whether access authorization has been configured for your account. For details, see Configuring Agency Authorization. If you have been authorized using access keys, clear the authorization and configure agency authorization.
2. In the navigation pane, choose Training Management > Training Jobs. The training job list is displayed by default.
3. Click Create Training Job. On the page that is displayed, configure parameters and click Next.
   • Created By: Custom algorithms
   • Boot Mode: Custom images
   • Image path: image created in Step 6 Uploading the Image to SWR.
   • Code Directory: directory where the boot script file is stored in OBS, for example, obs://test-modelarts/mindspore-gpu/resnet/. The training code is automatically downloaded to the ${MA_JOB_DIR}/resnet directory of the training container. resnet (customizable) is the last-level directory of the OBS path.
   • Boot Command: bash ${MA_JOB_DIR}/resnet/run_mpi.sh python ${MA_JOB_DIR}/resnet/train.py. resnet (customizable) is the last-level directory of the OBS path.
   • Training Input: Click Add Training Input. Enter data_path for the name, select the OBS path to the target dataset, for example, obs://test-modelarts/mindspore-gpu/cifar-10-batches-bin/, and set Obtained from to Hyperparameters.
   • Training Output: Click Add Training Output. Enter output_path for the name, select an OBS path for storing training outputs, for example, obs://test-modelarts/mindspore-gpu/output/, and set Obtained from to Hyperparameters and Predownload to No.
   • Hyperparameters: Click Add Hyperparameter and add the following hyperparameters:
     • run_distribute=True
     • device_num=1 (Set this parameter based on the number of GPUs in the instance flavors.)
     • device_target=GPU
     • epoch_size=2
   • Environment Variable: Click Add Environment Variable and add the environment variable MY_SSHD_PORT=38888.
   • Resource Pool: Select Public resource pools.
   • Resource Type: Select GPU.
   • Compute Nodes: 1 or 2
   • Persistent Log Saving: enabled
   • Job Log Path: OBS path to stored training logs, for example, obs://test-modelarts/mindspore-gpu/log/
4. Confirm the configurations of the training job and click Submit.
5. Wait until the training job is created.

   After you submit the job creation request, the system will automatically perform operations on the backend, such as downloading the container image and code directory and running the boot command. A training job requires a certain period of time for running. The duration ranges from dozens of minutes to several hours, varying depending on the service logic and selected resources. After the training job is executed, the log similar to the following is output.

   Figure 7 Run logs of training jobs with GPU specifications (one compute node)
   
   Figure 8 Run logs of training jobs with GPU specifications (two compute nodes)