scale_hdfs

Role Definition

• Role name: HDFS
• Definition:
  • HDFS is a network protocol used by Windows-based computers that allows systems within the same network to share files.
  • It allows computers connected to the same network or domain to access files from other local computers as easily as if they were on the computer's local hard drive.

Prerequisite

• CES should be enabled and configured.
• Atleast 1 node should be designated as hdfs node.
• Native hdfs service should be stopped, before usinf this role.
• Atleast one dedicated cesip required for hdfs role.
• Make sure, JAVA 1.8.0 version(java-<1.8.0>-openjdk-devel) should be installed on all the hdfs host nodes and JAVA_HOME exported. -1:

  yum installjava-1.8.0-openjdk-devel

  -2:

  vi ~/.bashrc

  export JAVA_HOME=/usr/lib/jvm/java-1.8.0-openjdk

  export PATH=$PATH:$JAVA_HOME/bin

Design

• Directory Structure:
  • Path: /ibm-spectrum-scale-install-infra/roles/scale_hdfs
  • Inside the scale_hdfs role, there are four more roles to enable microservice architecture:
    • Precheck: This role checks that all the prerequisites are satisfied before installing the hdfs rpms. It checks the following things:
      • Whether hdfs protocol is enabled or not.
      • Whether atleast one hdfs node is specified or not.
      • Whether supporting JAVA installed on all the hdfs nodes.
    • Node: This role installs the rpms required for hdfs protocol functionality. There are 3 methods to install hdfs rpms:
      • via local package, requires scale_install_localpkg_path variable to be set in main playbook.
      • via remote package, requires scale_install_remotepkg_path variable to be set in main playbook.
      • via yum repository, requires scale_install_callhome_repository_url variable to be set in main playbook.
    • Cluster: This role configure and enables the hdfs protocol.
    • Postcheck: This role checks if hdfs service is running or not.

Implementation

• Precheck

  • This role creates a list of hdfs nodes and checks if atleast one node is set as hdfs nodes.
  • It checks if all prerequisites are satisfied like supported JAVA installed and JAVA_HOME set(currently java-1.8.0-openjdk supported)

• Node

  • Installation via local package: rpms are on local machine
    • File name: install_local_pkg.yml
    • Path: /root/ibm-spectrum-scale-install-infra/roles/scale_hdfs/node/tasks
    • This playbook checks whether the package exists, checks for valid checksum, extracts all the rpms and creates a list ‘scale_install_all_rpms’ that contains all the rpms and dependencies required to install hdfs.
  • Installation via remote package: rpms are on remote machine
    • File name: install_remote_pkg.yml
    • Path: /root/ibm-spectrum-scale-install-infra/roles/scale_hdfs/node/tasks
    • This playbook checks whether the package exists, checks for valid checksum, extracts all the rpms and creates a list ‘scale_install_all_rpms’ that contains all the rpms and dependencies required to install hdfs.
  • Installation via repository: rpms are in yum repo
    • File name: install_repository.yml
    • Path: /root/ibm-spectrum-scale-install-infra/roles/scale_hdfs/node/tasks
    • This playbook configures the yum repo using the URL mentioned in variable scale_install_hdfs_repository_url and creates a list ‘scale_install_all_rpms’ that contains all the rpms and dependencies required to install hdfs.
  • The installation method is selected in the playbook ‘install.yml’ . Installation method depends on the variables defined.
    • If ‘scale_install_hdfs_repository_url’ is defined, then the installation method is repository.
    • If ‘scale_install_repository_url’ is undefined and ‘scale_install_remotepkg_path’ is defined, then the installation method is remote.
    • If ‘scale_install_repository_url’ is undefined and ‘scale_install_remotepkg_path’ is undefined and ‘scale_install_localpkg_path’ is defined, then the installation method is local.
  • Depending on the installation method, appropriate playbook is called for collecting the rpms and rpms are installed on all the nodes on which hdfs is enabled.

• Cluster

  • This role setup all hdfs required configurations.
  • To enable hdfs protocol, we run command mmces service enable hdfs on all the nodes where hdfs is enabled.

• Postcheck

  • This role verify namenodes and datanodes status.
  • This role uses command mmces service list to check if hdfs service is up and running.