In the early days of Hadoop 0.23 the thinking went that might be better off having a "mapreduce" user id for the hadoop mapreduce daemons. Since then folks have been complaining that:
- mapreduce doesn't get displayed by ps (it is too long)
- it is different from Hadoop 1.0 daemons packaging
- in a kerberized deployment it is confusing to have 2 types of principals
Personally, I'm fully convinced now that going back to 'mapred' as a user name is reasonable. The patch is also pretty trivial. I'll attach it and you guys can let me know if there's a reason we shouldn't apply it.