HDFS Requirements

Cluster mode pipelines that read from HDFS require the Cloudera distribution of Hadoop (CDH) or Hortonworks Data Platform (HDP). For a list of the supported CDH or HDP versions, see Available Stage Libraries in the Data Collector documentation.

Complete the following steps to configure a cluster mode pipeline to read from HDFS:

  1. Verify the installation of HDFS and YARN.
  2. Install Data Collector on a YARN gateway node.
  3. Grant the user defined in the user environment variable write permission on /user/$SDC_USER.
    The user environment variable defines the system user used to run Data Collector as a service. The file that defines the user environment variable depends on your operating system. For more information, see User and Group for Service Start in the Data Collector documentation.
    For example, say the user environment variable is defined as sdc and the cluster does not use Kerberos. Then you might use the following commands to create the directory and configure the necessary write permissions:
    $sudo -u hdfs hadoop fs -mkdir /user/sdc
    $sudo -u hdfs hadoop fs -chown sdc /user/sdc
  4. To enable Data Collector to submit YARN jobs, perform one of the following tasks:
    • On YARN, set the min.user.id to a value equal to or lower than the user ID associated with the Data Collector user ID, typically named "sdc".
    • On YARN, add the Data Collector user name, typically "sdc", to the allowed.system.users property.
    • After you create the pipeline, specify a Hadoop FS user in the Hadoop FS origin.

      For the Hadoop FS User property, enter a user with an ID that is higher than the min.user.id property, or with a user name that is listed in the allowed.system.users property.

  5. On YARN, verify that the Hadoop logging level is set to a severity of INFO or lower.
    YARN sets the Hadoop logging level to INFO by default. To change the logging level:
    1. Edit the log4j.properties file.
      By default, the file is located in the following directory:
    2. Set the log4j.rootLogger property to a severity of INFO or lower, such as DEBUG or TRACE.
  6. If YARN is configured to use Kerberos authentication, configure Data Collector to use Kerberos authentication.
    When you configure Kerberos authentication for Data Collector, you enable Data Collector to use Kerberos and define the principal and keytab.
    Important: For cluster pipelines, enter an absolute path to the keytab when configuring Data Collector. Standalone pipelines do not require an absolute path.
    Once enabled, Data Collector automatically uses the Kerberos principal and keytab to connect to any YARN cluster that uses Kerberos. For more information about enabling Kerberos authentication for Data Collector, see Kerberos Authentication in the Data Collector documentation.
  7. In the pipeline properties, on the General tab, set the Execution Mode property to Cluster Batch.
  8. On the Cluster tab, configure the following properties:
    Cluster Property Description
    Worker Java Options Additional Java properties for the pipeline. Separate properties with a space.

    The following properties are set by default.

    • XX:+UseConcMarkSweepGC and XX:+UseParNewGC are set to the Concurrent Mark Sweep (CMS) garbage collector.
    • Dlog4j.debug enables debug logging for log4j.

    Changing the default properties is not recommended.

    You can add any valid Java property.

    Launcher Env Configuration

    Additional configuration properties for the cluster launcher. Using simple or bulk edit mode, click the Add icon and define the property name and value.

    Worker Memory (MB) Maximum amount of memory allocated to each Data Collector worker in the cluster.

    Default is 1024 MB.

  9. In the pipeline, use the Hadoop FS origin for cluster mode.
    On the General tab of the origin, select the appropriate CDH or HDP stage library for cluster mode. For more information about the origin, see Hadoop FS (deprecated).
  10. If YARN is configured to use Kerberos authentication, in the origin, enable the Kerberos Authentication property on the Hadoop FS tab.