I am using 0.21.0 build. It seems like having different settings for properties in mapred-site.xml on different nodes within the cluster isn't supported? I want to set JVM flags for map/reduce JVMs differently on each of the data nodes. It looks like Hadoop is picking these options from the namenode's mapred-site.xml.
If this is something that Hadoop doesn't supports then how do folks set up Java heap requirements according to the memory available on each of the data nodes which could be different. Thanks, -Shrinivas