If you run the Dremio application as dremio
but use a keytab for a user named data_reader
, then YARN will fail to start with an error saying that the data_reader account does not have permission to “/user/dremio”.
When I dug through the code, I found that the hadoop-common library is defaulting the home directory/base directory based on the system property “user.name”, but if you try overriding this at Dremio startup (-Duser.name=data_reader
) you run into other issues around DB ownership.
If you run across this, here is the code change I made to work around it: