Kerberos Configuration Strategies for CDP

An overview of integration options when configuring Kerberos for Cloudera Data Platform (CDP).

Table 1. Kerberos Configuration Strategies for CDP
A. Direct integration of Cloudera Manager with AD/MIT/IPA KDC B. Keytab retrieval through “keytab retrieval script" C. Manual deployment / configuration of keytab for services
Brief description of functionality Cloudera Manager can create/delete service principals directly in the KDC as needed, and Cloudera Manager will retrieve the keytabs for the created principals automatically. Cloudera Manager will call an external script to retrieve a keytab for a given principal. Cloudera Manager will maintain keytab/kerberos related configuration for services with either the A or B column methods, but we will override the default location of the keytab with a “Safety Valve” option.

This is the default/recommended setup.

  • Cloudera Manager will store/track the available keytabs, and can create missing principals/keytabs after new services/role instances are added.
  • Cloudera Manager can recreate existing principals / keytabs if needed for some reason.
  • Where direct integration with AD / KDC is not feasible due to internal restrictions/policy.
  • Could be used integrate Cloudera Manager with unsupported KDC types.
Could be used where customizations are needed to the keytabs due to external factors.
Principal naming convention

Kerberos principals used by CDP are following this naming convention: service/fqdn@REALM.

Most services use a single username across the cluster (like “hdfs” or “hive”) for all roles of the service. The host part will vary based on the available hosts in the cluster, so a single service will likely need multiple principals, where the username part is the same, and the hostname will match one of the hosts where this service is running on. The default usernames used by the services can be found in Hadoop Users (user:group) and Kerberos Principals.

Scripts involved Cloudera Manager invokes these external shell scripts to perform credential management:

Base directory: /opt/cloudera/cm/bin

The “custom keytab retrieval script” will get a single principal name matching the above naming convention, and it should write the matching keytab to a directory also passed as an option to the script.

There is a basic sample script provided in Using a custom Kerberos keytab retrieval scrip.

Persistent storage of keytabs Keytabs that are successfully retrieved will be stored in the Cloudera Manager-backed database to be deployed and used on demand. None. Required keytabs should be stored on the hosts manually.
On-demand keytab merging

In case a service (role instance) requires multiple principals for operation, Cloudera Manager will merge the individual keytabs into a single combined keytab automatically.

For example: HDFS namenode uses both the “hdfs” and “HTTP” principal for operation. Cloudera Manager will create a merged keytab that contains both these principals.

None. In case a service (role instance) requires multiple principals for operation, keytabs provided for the services (role instance) are needed to be merged manually.
On-demand keytab deployment Cloudera Manager will automatically deploy the required configuration and keytab files into the process directory of the given role instance through the Cloudera Manager Agent when starting a service/role instance. None. All required keytabs will need to be present on the host(s) for a service before starting it. Cloudera Manager should be used to configure the location of these keytab files though.
Keytab filename

The keytab file & filename will be managed by Cloudera Manager. The keytab file will be deployed by the Cloudera Manager agent to the services process folder automatically.

The filenames used by Cloudera Manager can be found in Hadoop Users (user:group) and Kerberos Principals.

The name & location of the keytab file has to be specified manually, so it can be anything allowed by POSIX conventions.
Keytab file ownership and permissions

The “System User” of the respective service needs read-write access for the keytab file. Cloudera Manager will set the proper owner/group on the file with permission mode 600 (read-write for owner, no access for group/others)

Default owner/group and permissions can also be found in Hadoop Users (user:group) and Kerberos Principals.

The “System User” of the respective service needs read-write access for the keytab file. The proper permission and ownership of the file needs to be set manually.