Ranger controls access to various Hadoop services, so appropriate policies must be place to provide VectorH access to the services it requires.

VectorH needs full and exclusive access to all files under the /Actian directory in HDFS. Also, some of the Hadoop services require the VectorH user (actian) to have access to additional directories for housekeeping purposes.

The easiest way to ensure such access is to enable "federated mode" for Ranger in HDFS (in the Ambari console: HDFS, Configs, Advanced, Advanced ranger-hdfs-security, xasecure.add-hadoop-authorization=true) and letting HDFS access control default to the existing POSIX permissions.

When access control through Ranger policies is desired, you must define a policy to allow VectorH to access the appropriate locations in HDFS. In its simplest form, such a policy could look like this:

VectorH must have access to the queue that is configured in the VectorH YARN_AM_QUEUE configuration parameter, typically "default".

The easiest way to achieve this is to keep the YARN-ACL fallback enabled and let YARN access control default to the existing YARN ACL permissions. If strict Ranger access control is desired, you must disable the fallback (YARN, Configs, Advanced, Custom ranger-yarn-security, ranger.add-yarn-authorization=false), and then define a suitable policy to allow VectorH access to the queue.

The policy for YARN in Ranger that is automatically created upon install typically gives access to all queues to a given list of users. Simply adding the actian user to that list will work, but will give VectorH more rights than it strictly needs. Defining a separate policy to give the actian user access to only the indicated queue would be a better security policy.

In its simplest form, such a policy could look like this:

When running on a Kerberos-enabled Hadoop system, YARN may refuse user actian because it has by default an ID of 300, which is below the usual minimum allowed ID for Hadoop users. In this case, user actian should be added to the "allowed.system.users" parameter in container-executor.cfg on all nodes.