Prerequisites and limitations¶
This document only lists basic requirements for deploying the User Isolation Framework. Individual components may have additional requirements. Please refer to the detailed capabilities and reference architectures sections for more details.
This setup is mandatory for all deployments of UIF.
- ACL support must be enabled on the filesystem which hosts the DSS data directory
- You need root access to setup the User Isolation Framework
For each UNIX which will be impersonated by the DSS end-user (see Concepts for more details), the following requirements must be met:
- The user must have a valid UNIX account.
- The user must have a valid shell (ie, must be able to perform shell actions).
- The user must have a writable home directory on HDFS.
Each group of users in DSS should have a matching UNIX group of users locally.
Some isolation framework capabilities are available on:
- Cloudera CDH 5.9 and higher (as long as it meets the CDH distribution support notes)
- Hortonworks HDP 2.5 and higher (as long as it meets the HDP distribution support notes)
- EMR 5.19 and higher
User isolation is not supported on the other Hadoop distributions.
The following configuration is required on your Hadoop cluster:
- Kerberos security must be enabled.
- You need a keytab for the
dssuser, as described in Connecting to secure clusters
- You need administrator access to your Hadoop cluster to setup user isolation (to setup the DSS impersonation authorization)
User isolation is not supported on the other Hadoop distributions
If you want to leverage Hive:
- You must have Sentry or Ranger enabled
- HiveServer2 impersonation must be disabled (this is the default setting)
DSS can work with a restricted-access Hive metastore (ie, when only HiveServer2 can talk to the Metastore server), but due to limitations in Spark, a restricted-access metastore will disable the following features:
- Using a HiveContext in Spark code recipes (SQLContext remains available)
- Using table definitions from the Hive metastore in Spark code recipes (including SparkSQL)
- Running some visual recipes on Spark (since they require HiveContext-only features)
For each Hadoop which will be impersonated by the DSS end-user (see Concepts for more details), the following requirements must be met:
- The user must have a writable home directory on HDFS
Each group of users in DSS should have a matching group of users on Hadoop.
While manual configuration of all user accounts is fully possible, we recommend that you use a LDAP directory to have a unique source of truth for all users and group mappings, in DSS, on UNIX, and on Hadoop.
Migrating a DSS instance which was previously running without UIF is highly not recommended. We highly recommend starting with an empty DSS instance when setting up UIF.
“Downgrading” a DSS instance by disabling UIF is not supported.
When UIF is enabled, the following features are not available for end-users unless they have the “Write unsafe code” permission:
- Write custom partition dependency functions
- Write Python UDF in data preparation
For more information about the “Write unsafe code” permission, see Write unisolated code: details