Skip to content

Latest commit

 

History

History
299 lines (245 loc) · 15 KB

README.md

File metadata and controls

299 lines (245 loc) · 15 KB
Raw

How to Run a Deep Learning Job

Introduction

The system supports major deep learning frameworks, including CNTK and TensorFlow, etc. It also supports other type of workload through a customized docker image. Users need to prepare a config file and submit it for a job submission. This guide introduces the details of job submission.

Prerequisites

This guide assumes the system has already been deployed properly and a docker registry is available to store docker images.

Docker image

The system launches a deep learning job in one or more Docker containers. A Docker images is required in advance. The system provides a base Docker images with HDFS, CUDA and cuDNN support, based on which users can build their own custom Docker images.

To build a base Docker image, for example Dockerfile.build.base, run:

docker build -f Dockerfiles/Dockerfile.build.base -t qizhi.build.base:hadoop2.7.2-cuda8.0-cudnn6-devel-ubuntu16.04 Dockerfiles/

Then a custom docker image can be built based on it by adding FROM qizhi.build.base:hadoop2.7.2-cuda8.0-cudnn6-devel-ubuntu16.04 in the Dockerfile.

As an example, we customize a TensorFlow Docker image using Dockerfile.run.tensorflow:

docker build -f Dockerfiles/Dockerfile.run.tensorflow -t qizhi.run.tensorflow Dockerfiles/

Next, the built image is pushed to a docker registry for every node in the system to access that image:

docker tag qizhi.run.tensorflow your_docker_registry/qizhi.run.tensorflow
docker push your_docker_registry/qizhi.run.tensorflow

And the image is ready to serve. Note that above script assume the docker registry is deployed locally. Actual script can vary depending on the configuration of Docker registry.

Json config file for job submission

A json file describe detailed configuration required for a job submission. The detailed format is shown as below:

{
  "jobName":   String,
  "image":     String,
  "authFile":  String,
  "dataDir":   String,
  "outputDir": String,
  "codeDir":   String,
  "virtualCluster": String,
  "taskRoles": [
    {
      "name":       String,
      "taskNumber": Integer,
      "cpuNumber":  Integer,
      "memoryMB":   Integer,
      "gpuNumber":  Integer,
      "portList": [
        {
          "label": String,
          "beginAt": Integer,
          "portNumber": Integer
        }
      ],
      "command":    String
    }
  ],
  "gpuType": String,
  "killAllOnCompletedTaskNumber": Integer,
  "retryCount": Integer
}

Below please find the detailed explanation for each of the parameters in the config file:

Field Name Schema Description
jobName String in ^[A-Za-z0-9\-._~]+$ format, required Name for the job, need to be unique
image String, required URL pointing to the Docker image for all tasks in the job
authFile String, optional, HDFS URI Docker registry authentication file existing on HDFS
dataDir String, optional, HDFS URI Data directory existing on HDFS
outputDir String, optional, HDFS URI Output directory on HDFS, $QIZHI_DEFAULT_FS_URI/Output/$jobName will be used if not specified
codeDir String, optional, HDFS URI Code directory existing on HDFS
virtualCluster String, optional The virtual cluster job runs on. If omitted, the job will run on default virtual cluster
taskRoles List, required List of taskRole, one task role at least
taskRole.name String in ^[A-Za-z0-9._~]+$ format, required Name for the task role, need to be unique with other roles
taskRole.taskNumber Integer, required Number of tasks for the task role, no less than 1
taskRole.cpuNumber Integer, required CPU number for one task in the task role, no less than 1
taskRole.memoryMB Integer, required Memory for one task in the task role, no less than 100
taskRole.gpuNumber Integer, required GPU number for one task in the task role, no less than 0
taskRole.portList List, optional List of portType to use
taskRole.portType.label String in ^[A-Za-z0-9._~]+$ format, required Label name for the port type
taskRole.portType.beginAt Integer, required The port to begin with in the port type, 0 for random selection
taskRole.portType.portNumber Integer, required Number of ports for the specific type
taskRole.command String, required Executable command for tasks in the task role, can not be empty
gpuType String, optional Specify the GPU type to be used in the tasks. If omitted, the job will run on any gpu type
killAllOnCompletedTaskNumber Integer, optional Number of completed tasks to kill the entire job, no less than 0
retryCount Integer, optional Job retry count, no less than 0

If you're using a private Docker registry which needs authentication for image pull and is different from the registry used during deployment, please create an authentication file in the following format, upload it to HDFS and specify the path in authFile parameter in config file.

docker_registry_server
username
password

NOTE: If you're using a private registry at Docker Hub, you should use docker.io for docker_registry_server field in the authentication file.

Runtime environment

Each task in a job runs in one Docker container. For a multi-task job, one task might communicate with others. So a task need to be aware of other tasks' runtime information such as IP, port, etc. The system exposes such runtime information as environment variables to each task's Docker container. For mutual communication, user can write code in the container to access those runtime environment variables. Those environment variables can also be used in the job config file.

Below we show a complete list of environment variables accessible in a Docker container:

Environment Variable Name Description
QIZHI_WORK_DIR Working directory in Docker container
QIZHI_DEFAULT_FS_URI Default file system uri in QIZHI
QIZHI_JOB_NAME jobName in config file
QIZHI_JOB_VC_NAME The virtual cluster in which the job is running
QIZHI_USER_NAME User who submit the job
QIZHI_DATA_DIR dataDir in config file
QIZHI_OUTPUT_DIR outputDirin config file or the generated path if outputDir is not specified
QIZHI_CODE_DIR codeDir in config file
QIZHI_CURRENT_TASK_ROLE_NAME taskRole.name of current task role
QIZHI_CURRENT_TASK_ROLE_TASK_COUNT taskRole.taskNumber of current task role
QIZHI_CURRENT_TASK_ROLE_CPU_COUNT taskRole.cpuNumber of current task role
QIZHI_CURRENT_TASK_ROLE_MEM_MB taskRole.memoryMB of current task role
QIZHI_CURRENT_TASK_ROLE_GPU_COUNT taskRole.gpuNumber of current task role
QIZHI_CURRENT_TASK_ROLE_CURRENT_TASK_INDEX Index of current task in current task role, starting from 0
QIZHI_JOB_TASK_COUNT Total tasks' number in config file
QIZHI_JOB_TASK_ROLE_COUNT Total task roles' number in config file
QIZHI_JOB_TASK_ROLE_LIST Comma separated all task role names in config file
QIZHI_KILL_ALL_ON_COMPLETED_TASK_NUM killAllOnCompletedTaskNumber in config file
QIZHI_CONTAINER_HOST_IP Allocated ip for current docker container
QIZHI_CONTAINER_HOST_PORT_LIST Allocated port list for current docker container, in portLabel0:port0,port1,port2;portLabel1:port3,port4 format
QIZHI_CONTAINER_HOST__$type_PORT_LIST Allocated port list for portList.label == $type, comma separated port string
QIZHI_TASK_ROLE_$name_HOST_LIST Host list for QIZHI_TASK_ROLE_NAME == $name, comma separated ip:port string, sorted by current task index in task role. Each task role has a host list environment variable with the corresponding task role name

An example deep learning job

A distributed TensorFlow job is listed below as an example:

{
  "jobName": "tensorflow-distributed-jobguid",
  // customized tensorflow docker image with hdfs, cuda and cudnn support
  "image": "your_docker_registry/qizhi.run.tensorflow",
  // this example uses cifar10 dataset, which is available from
  // http://www.cs.toronto.edu/~kriz/cifar.html
  "dataDir": "$QIZHI_DEFAULT_FS_URI/path/tensorflow-distributed-jobguid/data",
  "outputDir": "$QIZHI_DEFAULT_FS_URI/path/tensorflow-distributed-jobguid/output",
  // this example uses code from tensorflow benchmark https://git.io/vF4wT
  "codeDir": "$QIZHI_DEFAULT_FS_URI/path/tensorflow-distributed-jobguid/code",
  "virtualCluster": "your_virtual_cluster",
  "taskRoles": [
    {
      "name": "ps_server",
      // use 2 ps servers in this job
      "taskNumber": 2,
      "cpuNumber": 2,
      "memoryMB": 8192,
      "gpuNumber": 0,
      "portList": [
        {
          "label": "http",
          "beginAt": 0,
          "portNumber": 1
        },
        {
          "label": "ssh",
          "beginAt": 0,
          "portNumber": 1
        }
      ],
      // run tf_cnn_benchmarks.py in code directory
      // please refer to https://www.tensorflow.org/performance/performance_models#executing_the_script for arguments' detail
      // if there's no `scipy` in the docker image, need to install it first
      "command": "pip --quiet install scipy && python code/tf_cnn_benchmarks.py --local_parameter_device=cpu --batch_size=32 --model=resnet20 --variable_update=parameter_server --data_dir=$QIZHI_DATA_DIR --data_name=cifar10 --train_dir=$QIZHI_OUTPUT_DIR --ps_hosts=$QIZHI_TASK_ROLE_ps_server_HOST_LIST --worker_hosts=$QIZHI_TASK_ROLE_worker_HOST_LIST --job_name=ps --task_index=$QIZHI_CURRENT_TASK_ROLE_CURRENT_TASK_INDEX"
    },
    {
      "name": "worker",
      // use 2 workers in this job
      "taskNumber": 2,
      "cpuNumber": 2,
      "memoryMB": 16384,
      "gpuNumber": 4,
      "portList": [
        {
          "label": "http",
          "beginAt": 0,
          "portNumber": 1
        },
        {
          "label": "ssh",
          "beginAt": 0,
          "portNumber": 1
        }
      ],
      "command": "pip --quiet install scipy && python code/tf_cnn_benchmarks.py --local_parameter_device=cpu --batch_size=32 --model=resnet20 --variable_update=parameter_server --data_dir=$QIZHI_DATA_DIR --data_name=cifar10 --train_dir=$QIZHI_OUTPUT_DIR --ps_hosts=$QIZHI_TASK_ROLE_ps_server_HOST_LIST --worker_hosts=$QIZHI_TASK_ROLE_worker_HOST_LIST --job_name=worker --task_index=$QIZHI_CURRENT_TASK_ROLE_CURRENT_TASK_INDEX"
    }
  ],
  // kill all 4 tasks when 2 worker tasks completed
  "killAllOnCompletedTaskNumber": 2,
  "retryCount": 0
}

More examples

Job submission

  1. Put the code and data on HDFS

    Use hdfs lib to upload your code and data to HDFS on the system.

  2. Prepare a job config file

    Prepare the config file for your job.

  3. Submit the job through web portal

    Open web portal in a browser, click "Submit Job" and upload your config file.

SSH Connection

You can ssh connect to a specified container either from outside or inside container.

SSH connect from outside

  1. Get job ssh connect info by invoking /api/v1/jobs/:jobName/ssh api or clicking the job detail page on webportal.

  2. Open a Bash shell terminal.

  3. Download the corresponding private key from HDFS. For example, with wget, you can execute below command line:

    wget http://host:port/webhdfs/v1/Container/userName/jobName/ssh/application_id/.ssh/application_id?op=OPEN -O application_id

  4. Use chmod command to set correct permission for the key file.

    chmod 400 application_id

  5. Use ssh command to connect into container. for example

    ssh -i application_id -p ssh_port root@container_ip

SSH connect inside containers

You can use ssh $QIZHI_CURRENT_TASK_ROLE_NAME-$QIZHI_CURRENT_TASK_ROLE_CURRENT_TASK_INDEX command to connect into another containers which belong to the same job. For example, if there are two taskRoles: master and worker, you can connect to worker-0 container directly with below command line:

ssh worker-0