Deployment of HDFS on a kubernete cluster.

The document describes the procedure to setup HDFS, either in high availability mode (multiple zookeepers and journal nodes, one active namenode and one standby namenode, multiple datanodes ), or with a single namenode across a cluster.

1. Configure HDFS configuration section on the cluster. You will need to specify: 1) the nodes, 2) block device used by HDFS.

  hdfs:
    partitions: /dev/sd[c|d]1
  kubelabels:
    hdfs: worker or <<hdfs_labels>> # nodes marked with hdfs_labels will be used for HDFS. 

In default, if the cluster has more than 1 infrastructure node, HDFS will be deployed in high availability(HA) mode. If the cluster has just 1 infrastructure node, HDFS will be deployed with single name node. Zookeeper and journal nodes only need to be deployed in HA mode, and can be skipped with single name node mode.

1. Format and partition device used by hdfs via:

     deploy.py hdfs create 
   

   This command may destroy all data on the target device, so you will be asked to manually confirm the operation.

If already formatted, you may just mount device by:

  deploy.py hdfs mount

The device can be unmounted via:

  deploy.py hdfs umount

1. Configure zookeeper and HDFS

     deploy.py docker push zookeeper
     deploy.py hdfs config
   

2. [High Available Cluster]Deploy HDFS zookeeper and journal node

     deploy.py kubernetes start zookeeper
     deploy.py kubernetes start hdfsjournal
   

   You may shutdown zookeeper and journal node via:

     deploy.py kubernetes stop hdfsjournal
     deploy.py kubernetes stop zookeeper
   

3. Format HDFS namenode/zookeeper Please wait for zookeeper and hdfsjournal pods to be launched, and then do:

     deploy.py kubernetes start hdfsformat
     [High Available Cluster] deploy.py kubernetes start hdfsstandby
     deploy.py kubernetes stop hdfsformat
     [High Available Cluster] deploy.py kubernetes stop hdfsstandby
   

   Command

     deploy.py --force kubernetes start hdfsformat
   

   will wipe out namenode information, and restart the cluster.

4. Deploy HDFS namenodes

     deploy.py kubernetes start hdfsnamenode
   

5. Deploy HDFS datanodes

     deploy.py kubernetes start hdfsdatanode
   

6. Mount HDFS volume

     deploy.py mount