1. Kevin Lyda
  2. intranode-balance



Intranode balancer

This tool requires that (at least) the node being balanced should have the DataNode process down. It's probably best to take down the cluster to stop hadoop from trying to make sure block replica minimums are met.

How to run

This is built specifically against version 1.0.3 of hadoop. It expects ../hadoop-common will contain a built version of hadoop. I'm not really good with Java so the build.xml file is kind of brittle this way.

The easiest way to run it is to just use ant:

ant run -Dargs=/path/to/hadoop/config

Do that as the hadoop user and it will move the blocks around.

There is no dry run option. If it fails in the middle of moving a block, it will be unhappy.