For creating a cluster, we want some idea about how the cluster should be created and cluster management.Here i am noting down some points for cluster creation and management process.
- Cloudera distribution is a good option to create a Hadoop cluster since it got a well structured repository and a well defined documentation set(Advanced user may go for the builds from Apache community).
- Cloudera Manager is designed to make administration of Hadoop simple and straight forward at any scale. With help of Cloudera manager, you can easily deploy and centrally operate the complete Hadoop stack. The application automates the installation process, reducing deployment time from week to minutes.
- Centos is the best option as OS since it’s developed on RHEL architecture and supports all RHEL add-ones.
- Yum install <packages> is a command that is used frequently for installing packages from remote repository. Yum install will pick the repository url from /etc/yum.repos.d, download the packages and install it in the machine. Normally yum will work in machine having internet access. But if we want to install packages in isolated environment, normal yum install will not work, because the remote repository may not be accessible in isolated environment. In that situation, we are creating a local yum repository.
- It’s better to turn off Graphical user experience in all the host machines, for efficient utilization of memory.
- For each installation add required environmental variables in /etc/bashrc file or /etc/profile for public access.
- For updating environment variables from /etc/bashrc file or /etc/profile files use ‘source’ command.
- Ensure sshd service is running in each node to make Secure Shell access active.
- Ensure IPtables service is stopped.
- Oracle jdk 1.6+ should be used (instead of open JDK) for JVM Process Status(JPS) which is used for displaying currently running Hadoop daemons.
- For tarball or source build installations ‘/opt’ location is preferred.
- Rebooting the Linux machine to change configurations is a bad practice and may negatively affect the overall cluster balance.
- For network connection issues restart network service other than rebooting the host machine.