Launching Applications Using Docker Containers

Security Warning

IMPORTANT Enabling this feature and running Docker containers in your cluster has security implications. Given Docker’s integration with many powerful kernel features, it is imperative that administrators understand Docker security before enabling this feature.

Overview

Docker combines an easy-to-use interface to Linux containers with easy-to-construct image files for those containers. In short, Docker enables users to bundle an application together with its preferred execution environment to be executed on a target machine. For more information about Docker, see their documentation.

The Linux Container Executor (LCE) allows the YARN NodeManager to launch YARN containers to run either directly on the host machine or inside Docker containers. The application requesting the resources can specify for each container how it should be executed. The LCE also provides enhanced security and is required when deploying a secure cluster. When the LCE launches a YARN container to execute in a Docker container, the application can specify the Docker image to be used.

Docker containers provide a custom execution environment in which the application’s code runs, isolated from the execution environment of the NodeManager and other applications. These containers can include special libraries needed by the application, and they can have different versions of native tools and libraries including Perl, Python, and Java. Docker containers can even run a different flavor of Linux than what is running on the NodeManager.

Docker for YARN provides both consistency (all YARN containers will have the same software environment) and isolation (no interference with whatever is installed on the physical machine).

Cluster Configuration

The LCE requires that container-executor binary be owned by root:hadoop and have 6050 permissions. In order to launch Docker containers, the Docker daemon must be running on all NodeManager hosts where Docker containers will be launched. The Docker client must also be installed on all NodeManager hosts where Docker containers will be launched and able to start Docker containers.

To prevent timeouts while starting jobs, any large Docker images to be used by an application should already be loaded in the Docker daemon’s cache on the NodeManager hosts. A simple way to load an image is by issuing a Docker pull request. For example:

    sudo docker pull library/openjdk:8

The following properties should be set in yarn-site.xml:

<configuration>
  <property>
    <name>yarn.nodemanager.container-executor.class</name>
    <value>org.apache.hadoop.yarn.server.nodemanager.LinuxContainerExecutor</value>
    <description>
      This is the container executor setting that ensures that all applications
      are started with the LinuxContainerExecutor.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.linux-container-executor.group</name>
    <value>hadoop</value>
    <description>
      The POSIX group of the NodeManager. It should match the setting in
      "container-executor.cfg". This configuration is required for validating
      the secure access of the container-executor binary.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.linux-container-executor.nonsecure-mode.limit-users</name>
    <value>false</value>
    <description>
      Whether all applications should be run as the NodeManager process' owner.
      When false, applications are launched instead as the application owner.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.allowed-runtimes</name>
    <value>default,docker</value>
    <description>
      Comma separated list of runtimes that are allowed when using
      LinuxContainerExecutor. The allowed values are default, docker, and
      javasandbox.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.type</name>
    <value></value>
    <description>
      Optional. Sets the default container runtime to use.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.image-name</name>
    <value></value>
    <description>
      Optional. Default docker image to be used when the docker runtime is
      selected.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.image-update</name>
    <value>false</value>
    <description>
      Optional. Default option to decide whether to pull the latest image
      or not.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.allowed-container-networks</name>
    <value>host,none,bridge</value>
    <description>
      Optional. A comma-separated set of networks allowed when launching
      containers. Valid values are determined by Docker networks available from
      `docker network ls`
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.default-container-network</name>
    <value>host</value>
    <description>
      The network used when launching Docker containers when no
      network is specified in the request. This network must be one of the
      (configurable) set of allowed container networks.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.host-pid-namespace.allowed</name>
    <value>false</value>
    <description>
      Optional. Whether containers are allowed to use the host PID namespace.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.privileged-containers.allowed</name>
    <value>false</value>
    <description>
      Optional. Whether applications are allowed to run in privileged
      containers. Privileged containers are granted the complete set of
      capabilities and are not subject to the limitations imposed by the device
      cgroup controller. In other words, privileged containers can do almost
      everything that the host can do. Use with extreme care.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.delayed-removal.allowed</name>
    <value>false</value>
    <description>
      Optional. Whether or not users are allowed to request that Docker
      containers honor the debug deletion delay. This is useful for
      troubleshooting Docker container related launch failures.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.stop.grace-period</name>
    <value>10</value>
    <description>
      Optional. A configurable value to pass to the Docker Stop command. This
      value defines the number of seconds between the docker stop command sending
      a SIGTERM and a SIGKILL.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.privileged-containers.acl</name>
    <value></value>
    <description>
      Optional. A comma-separated list of users who are allowed to request
      privileged containers if privileged containers are allowed.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.capabilities</name>
    <value>CHOWN,DAC_OVERRIDE,FSETID,FOWNER,MKNOD,NET_RAW,SETGID,SETUID,SETFCAP,SETPCAP,NET_BIND_SERVICE,SYS_CHROOT,KILL,AUDIT_WRITE</value>
    <description>
      Optional. This configuration setting determines the capabilities
      assigned to docker containers when they are launched. While these may not
      be case-sensitive from a docker perspective, it is best to keep these
      uppercase. To run without any capabilites, set this value to
      "none" or "NONE"
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.enable-userremapping.allowed</name>
    <value>true</value>
    <description>
      Optional. Whether docker containers are run with the UID and GID of the
      calling user.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.userremapping-uid-threshold</name>
    <value>1</value>
    <description>
      Optional. The minimum acceptable UID for a remapped user. Users with UIDs
      lower than this value will not be allowed to launch containers when user
      remapping is enabled.
    </description>
  </property>

  <property>
    <name>yarn.nodemanager.runtime.linux.docker.userremapping-gid-threshold</name>
    <value>1</value>
    <description>
      Optional. The minimum acceptable GID for a remapped user. Users belonging
      to any group with a GID lower than this value will not be allowed to
      launch containers when user remapping is enabled.
    </description>
  </property>

</configuration>

In addition, a container-executor.cfg file must exist and contain settings for the container executor. The file must be owned by root with permissions 0400. The format of the file is the standard Java properties file format, for example

`key=value`

The following properties are required to enable Docker support:

The container-executor.cfg must contain a section to determine the capabilities that containers are allowed. It contains the following properties:

Please note that if you wish to run Docker containers that require access to the YARN local directories, you must add them to the docker.allowed.rw-mounts list.

In addition, containers are not permitted to mount any parent of the container-executor.cfg directory in read-write mode.

The following properties are optional:

Part of a container-executor.cfg which allows Docker containers to be launched is below:

yarn.nodemanager.linux-container-executor.group=yarn
[docker]
  module.enabled=true
  docker.privileged-containers.enabled=true
  docker.privileged-containers.registries=local
  docker.trusted.registries=centos
  docker.allowed.capabilities=SYS_CHROOT,MKNOD,SETFCAP,SETPCAP,FSETID,CHOWN,AUDIT_WRITE,SETGID,NET_RAW,FOWNER,SETUID,DAC_OVERRIDE,KILL,NET_BIND_SERVICE
  docker.allowed.networks=bridge,host,none
  docker.allowed.ro-mounts=/sys/fs/cgroup
  docker.allowed.rw-mounts=/var/hadoop/yarn/local-dir,/var/hadoop/yarn/log-dir

Docker Image Requirements

In order to work with YARN, there are two requirements for Docker images.

First, the Docker container will be explicitly launched with the application owner as the container user. If the application owner is not a valid user in the Docker image, the application will fail. The container user is specified by the user’s UID. If the user’s UID is different between the NodeManager host and the Docker image, the container may be launched as the wrong user or may fail to launch because the UID does not exist. See User Management in Docker Container section for more details.

Second, the Docker image must have whatever is expected by the application in order to execute. In the case of Hadoop (MapReduce or Spark), the Docker image must contain the JRE and Hadoop libraries and have the necessary environment variables set: JAVA_HOME, HADOOP_COMMON_PATH, HADOOP_HDFS_HOME, HADOOP_MAPRED_HOME, HADOOP_YARN_HOME, and HADOOP_CONF_DIR. Note that the Java and Hadoop component versions available in the Docker image must be compatible with what’s installed on the cluster and in any other Docker images being used for other tasks of the same job. Otherwise the Hadoop components started in the Docker container may be unable to communicate with external Hadoop components.

If a Docker image has a command set, the behavior will depend on whether the YARN_CONTAINER_RUNTIME_DOCKER_RUN_OVERRIDE_DISABLE is set to true. If so, the command will be overridden when LCE launches the image with YARN’s container launch script.

If a Docker image has an entry point set and YARN_CONTAINER_RUNTIME_DOCKER_RUN_OVERRIDE_DISABLE is set to true, launch_command will be passed to ENTRYPOINT program as CMD parameters in Docker. The format of launch_command looks like: param1,param2 and this translates to CMD [ “param1”,“param2” ] in Docker.

If an application requests a Docker image that has not already been loaded by the Docker daemon on the host where it is to execute, the Docker daemon will implicitly perform a Docker pull command. Both MapReduce and Spark assume that tasks which take more that 10 minutes to report progress have stalled, so specifying a large Docker image may cause the application to fail.

CGroups configuration Requirements

The Docker plugin utilizes cgroups to limit resource usage of individual containers. Since launched containers belong to YARN, the command line option --cgroup-parent is used to define the appropriate control group.

Docker supports two different cgroups driver: cgroupfs and systemd. Note that only cgroupfs is supported - attempt to launch a Docker container with systemd results in the following, similar error message:

Container id: container_1561638268473_0006_01_000002
Exit code: 7
Exception message: Launch container failed
Shell error output: /usr/bin/docker-current: Error response from daemon: cgroup-parent for systemd cgroup should be a valid slice named as "xxx.slice".
See '/usr/bin/docker-current run --help'.
Shell output: main : command provided 4

This means you have to reconfigure the Docker deamon on each host where systemd driver is used.

Depending on what OS Hadoop is running on, reconfiguration might require different steps. However, if systemd was chosen for cgroups driver, it is likely that the systemctl command is available on the system.

Check the ExecStart property of the Docker daemon:

~$ systemctl show --no-pager --property=ExecStart docker.service
ExecStart={ path=/usr/bin/dockerd-current ; argv[]=/usr/bin/dockerd-current --add-runtime
docker-runc=/usr/libexec/docker/docker-runc-current --default-runtime=docker-runc --exec-opt native.cgroupdriver=systemd
--userland-proxy-path=/usr/libexec/docker/docker-proxy-current
--init-path=/usr/libexec/docker/docker-init-current
--seccomp-profile=/etc/docker/seccomp.json
$OPTIONS $DOCKER_STORAGE_OPTIONS $DOCKER_NETWORK_OPTIONS $ADD_REGISTRY $BLOCK_REGISTRY $INSECURE_REGISTRY $REGISTRIES ;
ignore_errors=no ; start_time=[n/a] ; stop_time=[n/a] ; pid=0 ; code=(null) ; status=0/0 }

This example shows that the native.cgroupdriver is systemd. You have to modify that in the unit file of the daemon.

~$ sudo systemctl edit --full docker.service

This brings up the whole configuration for editing. Just replace the systemd string to cgroupfs. Save the changes and restart both the systemd and Docker daemon:

~$ sudo systemctl daemon-reload
~$ sudo systemctl restart docker.service

Application Submission

Before attempting to launch a Docker container, make sure that the LCE configuration is working for applications requesting regular YARN containers. If after enabling the LCE one or more NodeManagers fail to start, the cause is most likely that the ownership and/or permissions on the container-executor binary are incorrect. Check the logs to confirm.

In order to run an application in a Docker container, set the following environment variables in the application’s environment:

The first two are required. The remainder can be set as needed. While controlling the container type through environment variables is somewhat less than ideal, it allows applications with no awareness of YARN’s Docker support (such as MapReduce and Spark) to nonetheless take advantage of it through their support for configuring the application environment.

Once an application has been submitted to be launched in a Docker container, the application will behave exactly as any other YARN application. Logs will be aggregated and stored in the relevant history server. The application life cycle will be the same as for a non-Docker application.

Using Docker Bind Mounted Volumes

WARNING Care should be taken when enabling this feature. Enabling access to directories such as, but not limited to, /, /etc, /run, or /home is not advisable and can result in containers negatively impacting the host or leaking sensitive information. WARNING

Files and directories from the host are commonly needed within the Docker containers, which Docker provides through volumes. Examples include localized resources, Apache Hadoop binaries, and sockets. To facilitate this need, YARN-6623 added the ability for administrators to set a whitelist of host directories that are allowed to be bind mounted as volumes into containers. YARN-5534 added the ability for users to supply a list of mounts that will be mounted into the containers, if allowed by the administrative whitelist.

In order to make use of this feature, the following must be configured.

• The administrator must define the volume whitelist in container-executor.cfg by setting docker.allowed.ro-mounts and docker.allowed.rw-mounts to the list of parent directories that are allowed to be mounted.
• The application submitter requests the required volumes at application submission time using the YARN_CONTAINER_RUNTIME_DOCKER_MOUNTS environment variable.

The administrator supplied whitelist is defined as a comma separated list of directories that are allowed to be mounted into containers. The source directory supplied by the user must either match or be a child of the specified directory.

The user supplied mount list is defined as a comma separated list in the form source:destination or source:destination:mode. The source is the file or directory on the host. The destination is the path within the container where the source will be bind mounted. The mode defines the mode the user expects for the mount, which can be ro (read-only) or rw (read-write). If not specified, rw is assumed. The mode may also include a bind propagation option (shared, rshared, slave, rslave, private, or rprivate). In that case, the mode should be of the form option, rw+option, or ro+option.

The following example outlines how to use this feature to mount the commonly needed /sys/fs/cgroup directory into the container running on YARN.

The administrator sets docker.allowed.ro-mounts in container-executor.cfg to “/sys/fs/cgroup”. Applications can now request that “/sys/fs/cgroup” be mounted from the host into the container in read-only mode.

At application submission time, the YARN_CONTAINER_RUNTIME_DOCKER_MOUNTS environment variable can then be set to request this mount. In this example, the environment variable would be set to “/sys/fs/cgroup:/sys/fs/cgroup:ro”. The destination path is not restricted, “/sys/fs/cgroup:/cgroup:ro” would also be valid given the example admin whitelist.

User Management in Docker Container

YARN’s Docker container support launches container processes using the uid:gid identity of the user, as defined on the NodeManager host. User and group name mismatches between the NodeManager host and container can lead to permission issues, failed container launches, or even security holes. Centralizing user and group management for both hosts and containers greatly reduces these risks. When running containerized applications on YARN, it is necessary to understand which uid:gid pair will be used to launch the container’s process.

As an example of what is meant by uid:gid pair, consider the following. By default, in non-secure mode, YARN will launch processes as the user nobody (see the table at the bottom of Using CGroups with YARN for how the run as user is determined in non-secure mode). On CentOS based systems, the nobody user’s uid is 99 and the nobody group is 99. As a result, YARN will call docker run with --user 99:99. If the nobody user does not have the uid 99 in the container, the launch may fail or have unexpected results.

One exception to this rule is the use of Privileged Docker containers. Privileged containers will not set the uid:gid pair when launching the container and will honor the USER or GROUP entries in the Dockerfile. This allows running privileged containers as any user which has security implications. Please understand these implications before enabling Privileged Docker containers.

There are many ways to address user and group management. Docker, by default, will authenticate users against /etc/passwd (and /etc/shadow) within the container. Using the default /etc/passwd supplied in the Docker image is unlikely to contain the appropriate user entries and will result in launch failures. It is highly recommended to centralize user and group management. Several approaches to user and group management are outlined below.

usermod -u 99 nobody
groupmod -g 99 nobody

application -> libpam -> pam_authenticate -> pam_unix.so -> /etc/passwd

application -> libpam -> pam_authenticate -> pam_sss.so -> SSSD -> pam_unix.so -> /etc/passwd

-v /var/lib/sss/pipes:/var/lib/sss/pipes:rw

Privileged Container Security Consideration

Privileged docker container can interact with host system devices. This can cause harm to host operating system without proper care. In order to mitigate risk of allowing privileged container to run on Hadoop cluster, we implemented a controlled process to sandbox unauthorized privileged docker images.

The default behavior disallows any privileged docker containers. Privileged docker is only allowed with ENTRYPOINT enabled docker image, and docker.privileged-containers.enabled is set to enabled. Docker image can run with root privileges in the docker container, but access to host level devices are disabled. This allows developer and tester to run docker images from internet with some restrictions to prevent harm to host operating system.

When docker images have been certified by developers and testers to be trustworthy. The trusted image can be promoted to trusted docker registry. System administrator can define docker.trusted.registries, and setup private docker registry server to promote trusted images. System administrator may choose to allow official docker images from Docker Hub to be part of trusted registries. “library” is the name to use for trusting official docker images. Container-executor.cfg example:

[docker]
  docker.privileged-containers.enabled=true
  docker.trusted.registries=library

Fine grained access control can also be defined using docker.privileged-containers.registries to allow only a subset of Docker images to run as privileged containers. If docker.privileged-containers.registries is not defined, YARN will fall back to use docker.trusted.registries as access control for privileged Docker images. Fine grained access control example:

[docker]
  docker.privileged-containers.enabled=true
  docker.privileged-containers.registries=local/centos:latest
  docker.trusted.registries=library

In development environment, local images can be tagged with a repository name prefix to enable trust. The recommendation of choosing a repository name is using a local hostname and port number to prevent accidentially pulling docker images from Docker Hub or use reserved Docker Hub keyword: “local”. Docker run will look for docker images on Docker Hub, if the image does not exist locally. Using a local hostname and port in image name can prevent accidental pulling of canonical images from docker hub. Example of tagging image with localhost:5000 as trusted registry:

docker tag centos:latest localhost:5000/centos:latest

Let’s say you have an Ubuntu-based image with some changes in the local repository and you wish to use it. The following example tags the local_ubuntu image:

docker tag local_ubuntu local/ubuntu:latest

Next, you have to add local to docker.trusted.registries. The image can be referenced by using local/ubuntu.

Trusted images are allowed to mount external devices such as HDFS via NFS gateway, or host level Hadoop configuration. If system administrators allow writing to external volumes using docker.allow.rw-mounts directive, privileged docker container can have full control of host level files in the predefined volumes.

For YARN Service HTTPD example, container-executor.cfg must define centos docker registry to be trusted for the example to run.

Container Reacquisition Requirements

On restart, the NodeManager, as part of the NodeManager’s recovery process, will validate that a container is still running by checking for the existence of the container’s PID directory in the /proc filesystem. For security purposes, operating system administrator may enable the hidepid mount option for the /proc filesystem. If the hidepid option is enabled, the yarn user’s primary group must be whitelisted by setting the gid mount flag similar to below. Without the yarn user’s primary group whitelisted, container reacquisition will fail and the container will be killed on NodeManager restart.

proc     /proc     proc     nosuid,nodev,noexec,hidepid=2,gid=yarn     0 0

Connecting to a Docker Trusted Registry

The Docker client command will draw its configuration from the default location, which is $HOME/.docker/config.json on the NodeManager host. The Docker configuration is where secure repository credentials are stored, so use of the LCE with secure Docker repos is discouraged using this method.

YARN-5428 added support to Distributed Shell for securely supplying the Docker client configuration. See the Distributed Shell help for usage. Support for additional frameworks is planned.

As a work-around, you may manually log the Docker daemon on every NodeManager host into the secure repo using the Docker login command:

  docker login [OPTIONS] [SERVER]

  Register or log in to a Docker registry server, if no server is specified
  "https://index.docker.io/v1/" is the default.

  -e, --email=""       Email
  -p, --password=""    Password
  -u, --username=""    Username

Note that this approach means that all users will have access to the secure repo.

Hadoop integrates with Docker Trusted Registry via YARN service API. Docker registry can store Docker images on HDFS, S3 or external storage using CSI driver.

    <property>
      <name>nfs.exports.allowed.hosts</name>
      <value>* rw</value>
    </property>

    <property>
      <name>nfs.file.dump.dir</name>
      <value>/tmp/.hdfs-nfs</value>
    </property>

    <property>
      <name>nfs.kerberos.principal</name>
      <value>nfs/_HOST@EXAMPLE.COM</value>
    </property>

    <property>
      <name>nfs.keytab.file</name>
      <value>/etc/security/keytabs/nfs.service.keytab</value>
    </property>

$ $HADOOP_HOME/bin/hdfs --daemon start nfs3

# mount -t nfs -o vers=3,proto=tcp,nolock,noacl,sync $DN_IP:/ /hdfs

[docker]
  docker.privileged-containers.enabled=true
  docker.trusted.registries=library,registry.docker-registry.registry.example.com:5000
  docker.allowed.rw-mounts=/tmp,/usr/local/hadoop/logs,/hdfs

{
  "name": "docker-registry",
  "version": "1.0",
  "kerberos_principal" : {
    "principal_name" : "registry/_HOST@EXAMPLE.COM",
    "keytab" : "file:///etc/security/keytabs/registry.service.keytab"
  },
  "components" :
  [
    {
      "name": "registry",
      "number_of_containers": 1,
      "artifact": {
        "id": "registry:latest",
        "type": "DOCKER"
      },
      "resource": {
        "cpus": 1,
        "memory": "256"
      },
      "run_privileged_container": true,
      "configuration": {
        "env": {
          "YARN_CONTAINER_RUNTIME_DOCKER_RUN_OVERRIDE_DISABLE":"true",
          "YARN_CONTAINER_RUNTIME_DOCKER_MOUNTS":"/hdfs/apps/docker/registry:/var/lib/registry"
        },
        "properties": {
          "docker.network": "host"
        }
      }
    }
  ]
}

yarn app -launch docker-registry /tmp/registry.json

registry.docker-registry.$USER.$DOMAIN:5000

version: 0.1
log:
    fields:
        service: registry
http:
    addr: :5000
storage:
    cache:
        blobdescriptor: inmemory
    s3:
        accesskey: #AWS_KEY#
        secretkey: #AWS_SECRET#
        region: #AWS_REGION#
        bucket: #AWS_BUCKET#
        encrypt: #ENCRYPT#
        secure:  #SECURE#
        chunksize: 5242880
        multipartcopychunksize: 33554432
        multipartcopymaxconcurrency: 100
        multipartcopythresholdsize: 33554432
        rootdirectory: #STORAGE_PATH#

{
  "name": "docker-registry",
  "version": "1.0",
  "kerberos_principal" : {
    "principal_name" : "registry/_HOST@EXAMPLE.COM",
    "keytab" : "file:///etc/security/keytabs/registry.service.keytab"
  },
  "components" :
  [
    {
      "name": "registry",
      "number_of_containers": 1,
      "artifact": {
        "id": "registry:latest",
        "type": "DOCKER"
      },
      "resource": {
        "cpus": 1,
        "memory": "256"
      },
      "run_privileged_container": true,
      "configuration": {
        "env": {
          "YARN_CONTAINER_RUNTIME_DOCKER_RUN_OVERRIDE_DISABLE":"true",
          "YARN_CONTAINER_RUNTIME_DOCKER_MOUNTS":"<path to config.yml>:/etc/docker/registry/config.yml",
        },
        "properties": {
          "docker.network": "host"
        }
      }
    }
  ]
}

Example: MapReduce

This example assumes that Hadoop is installed to /usr/local/hadoop.

Additionally, docker.allowed.ro-mounts in container-executor.cfg has been updated to include the directories: /usr/local/hadoop,/etc/passwd,/etc/group.

To submit the pi job to run in Docker containers, run the following commands:

  HADOOP_HOME=/usr/local/hadoop
  YARN_EXAMPLES_JAR=$HADOOP_HOME/share/hadoop/mapreduce/hadoop-mapreduce-examples-*.jar
  MOUNTS="$HADOOP_HOME:$HADOOP_HOME:ro,/etc/passwd:/etc/passwd:ro,/etc/group:/etc/group:ro"
  IMAGE_ID="library/openjdk:8"

  export YARN_CONTAINER_RUNTIME_TYPE=docker
  export YARN_CONTAINER_RUNTIME_DOCKER_IMAGE=$IMAGE_ID
  export YARN_CONTAINER_RUNTIME_DOCKER_MOUNTS=$MOUNTS

  yarn jar $YARN_EXAMPLES_JAR pi \
    -Dmapreduce.map.env.YARN_CONTAINER_RUNTIME_TYPE=docker \
    -Dmapreduce.map.env.YARN_CONTAINER_RUNTIME_DOCKER_MOUNTS=$MOUNTS \
    -Dmapreduce.map.env.YARN_CONTAINER_RUNTIME_DOCKER_IMAGE=$IMAGE_ID \
    -Dmapreduce.reduce.env.YARN_CONTAINER_RUNTIME_TYPE=docker \
    -Dmapreduce.reduce.env.YARN_CONTAINER_RUNTIME_DOCKER_MOUNTS=$MOUNTS \
    -Dmapreduce.reduce.env.YARN_CONTAINER_RUNTIME_DOCKER_IMAGE=$IMAGE_ID \
    1 40000

Note that the application master, map tasks, and reduce tasks are configured independently. In this example, we are using the openjdk:8 image for all three.

Example: Spark

This example assumes that Hadoop is installed to /usr/local/hadoop and Spark is installed to /usr/local/spark.

Additionally, docker.allowed.ro-mounts in container-executor.cfg has been updated to include the directories: /usr/local/hadoop,/etc/passwd,/etc/group.

To run a Spark shell in Docker containers, run the following command:

  HADOOP_HOME=/usr/local/hadoop
  SPARK_HOME=/usr/local/spark
  MOUNTS="$HADOOP_HOME:$HADOOP_HOME:ro,/etc/passwd:/etc/passwd:ro,/etc/group:/etc/group:ro"
  IMAGE_ID="library/openjdk:8"

  $SPARK_HOME/bin/spark-shell --master yarn \
    --conf spark.yarn.appMasterEnv.YARN_CONTAINER_RUNTIME_TYPE=docker \
    --conf spark.yarn.appMasterEnv.YARN_CONTAINER_RUNTIME_DOCKER_IMAGE=$IMAGE_ID \
    --conf spark.yarn.appMasterEnv.YARN_CONTAINER_RUNTIME_DOCKER_MOUNTS=$MOUNTS \
    --conf spark.executorEnv.YARN_CONTAINER_RUNTIME_TYPE=docker \
    --conf spark.executorEnv.YARN_CONTAINER_RUNTIME_DOCKER_IMAGE=$IMAGE_ID \
    --conf spark.executorEnv.YARN_CONTAINER_RUNTIME_DOCKER_MOUNTS=$MOUNTS

Note that the application master and executors are configured independently. In this example, we are using the openjdk:8 image for both.

When using remote container registry, the YARN_CONTAINER_RUNTIME_DOCKER_CLIENT_CONFIG must reference the config.json file containing the credentials used to authenticate.

DOCKER_IMAGE_NAME=hadoop-docker
DOCKER_CLIENT_CONFIG=hdfs:///user/hadoop/config.json
spark-submit --master yarn \
--deploy-mode cluster \
--conf spark.executorEnv.YARN_CONTAINER_RUNTIME_TYPE=docker \
--conf spark.executorEnv.YARN_CONTAINER_RUNTIME_DOCKER_IMAGE=$DOCKER_IMAGE_NAME \
--conf spark.executorEnv.YARN_CONTAINER_RUNTIME_DOCKER_CLIENT_CONFIG=$DOCKER_CLIENT_CONFIG \
--conf spark.yarn.appMasterEnv.YARN_CONTAINER_RUNTIME_TYPE=docker \
--conf spark.yarn.appMasterEnv.YARN_CONTAINER_RUNTIME_DOCKER_IMAGE=$DOCKER_IMAGE_NAME \
--conf spark.yarn.appMasterEnv.YARN_CONTAINER_RUNTIME_DOCKER_CLIENT_CONFIG=$DOCKER_CLIENT_CONFIG \
sparkR.R

Docker Container ENTRYPOINT Support

When Docker support was introduced to Hadoop 2.x, the platform was designed to run existing Hadoop programs inside Docker container. Log redirection and environment setup are integrated with Node Manager. In Hadoop 3.x, Hadoop Docker support extends beyond running Hadoop workload, and support Docker container in Docker native form using ENTRYPOINT from dockerfile. Application can decide to support YARN mode as default or Docker mode as default by defining YARN_CONTAINER_RUNTIME_DOCKER_RUN_OVERRIDE_DISABLE environment variable. System administrator can also set as default setting for the cluster to make ENTRY_POINT as default mode of operation.

In yarn-site.xml, add YARN_CONTAINER_RUNTIME_DOCKER_RUN_OVERRIDE_DISABLE to node manager environment white list:

<property>
        <name>yarn.nodemanager.env-whitelist</name>
        <value>JAVA_HOME,HADOOP_COMMON_HOME,HADOOP_HDFS_HOME,HADOOP_CONF_DIR,HADOOP_YARN_HOME,HADOOP_HOME,PATH,LANG,TZ,HADOOP_MAPRED_HOME,YARN_CONTAINER_RUNTIME_DOCKER_RUN_OVERRIDE_DISABLE</value>
</property>

In yarn-env.sh, define:

export YARN_CONTAINER_RUNTIME_DOCKER_RUN_OVERRIDE_DISABLE=true

Requirements when not using ENTRYPOINT (YARN mode)

There are two requirements when ENTRYPOINT is not used:

1. /bin/bash must be available inside the image. This is generally true, however, tiny Docker images (eg. ones which use busybox for shell commands) might not have bash installed. In this case, the following error is displayed:

   Container id: container_1561638268473_0015_01_000002
   Exit code: 7
   Exception message: Launch container failed
   Shell error output: /usr/bin/docker-current: Error response from daemon: oci runtime error: container_linux.go:235: starting container process caused "exec: \"bash\": executable file not found in $PATH".
   Shell output: main : command provided 4
   

2. find command must also be available inside the image. Not having find causes this error:

   Container exited with a non-zero exit code 127. Error file: prelaunch.err.
   Last 4096 bytes of prelaunch.err :
   /tmp/hadoop-systest/nm-local-dir/usercache/hadoopuser/appcache/application_1561638268473_0017/container_1561638268473_0017_01_000002/launch_container.sh: line 44: find: command not found
   

Docker Container YARN SysFS Support

YARN SysFS is a pseudo file system provided by the YARN framework that exports information about clustering information to Docker container. Cluster information is exported to /hadoop/yarn/sysfs path. This API allows application developer to obtain clustering information without external service dependencies. Custom application master can populate cluster information by calling node manager REST API. YARN service framework automatically populates cluster information to /hadoop/yarn/sysfs/app.json. For more information about YARN service, see: YARN Service.

Docker Container Service Mode

Docker Container Service Mode runs the container as defined by the image but does not set the user (–user and –group-add). This mode is disabled by default. The administrator sets docker.service-mode.enabled to true in container-executor.cfg under docker section to enable.

Part of a container-executor.cfg which allows docker service mode is below:

yarn.nodemanager.linux-container-executor.group=yarn
[docker]
  module.enabled=true
  docker.privileged-containers.enabled=true
  docker.service-mode.enabled=true

Application User can enable or disable service mode at job level by exporting environment variable YARN_CONTAINER_RUNTIME_DOCKER_SERVICE_MODE in the application’s environment with value true or false respectively.