Sentry Policy File Authorization

Policy File

The sections that follow contain notes on creating and maintaining the policy file, and using URIs to load external data and JARs.

Storing the Policy File

Considerations for storing the policy file(s) in HDFS include:

  1. Replication count - Because the file is read for each query in Hive and read once every five minutes by all Impala daemons, you should increase this value; since it is a small file, setting the replication count equal to the number of client nodes in the cluster is reasonable.
  2. Updating the file - Updates to the file are reflected immediately, so you should write them to a temporary copy of the file first, and then replace the existing file with the temporary one after all the updates are complete. This avoids race conditions caused by reads on an incomplete file.

Defining Roles

Keep in mind that role definitions are not cumulative; the definition that is further down in the file replaces the older one. For example, the following results in role1 having privilege2, not privilege1 and privilege2.
role1 = privilege1
role1 = privilege2
Role names are scoped to a specific file. For example, if you give role1 the ALL privilege on db1 in the global policy file and give role1 ALL on db2 in the per-db db2 policy file, the user will be given both privileges.

URIs

Any command which references a URI such as CREATE TABLE EXTERNAL, LOAD, IMPORT, EXPORT, and more, in addition to CREATE TEMPORARY FUNCTION requires the URI privilege. This is an important security control because without this users could simply create an external table over an existing table they do not have access to and bypass Sentry.

URIs must start with either hdfs:// or file://. If a URI starts with anything else, it will cause an exception and the policy file will be invalid.

When defining URIs for HDFS, you must also specify the NameNode. For example:
data_read = server=server1->uri=file:///path/to/dir,\
server=server1->uri=hdfs://namenode:port/path/to/dir

Loading Data

Data can be loaded using a landing skid, either in HDFS or using a local/NFS directory where HiveServer2/Impala run. The following privileges can be used to grant a role access to a loading skid:
  • Load data from a local/NFS directory:
    server=server1->uri=file:///path/to/nfs/local/to/nfs
  • Load data from HDFS (MapReduce, Pig, and so on):
    server=server1->uri=hdfs://ha-nn-uri/data/landing-skid

In addition to the privilege in Sentry, the hive or impala user will require the appropriate file permissions to access the data being loaded. Groups can be used for this purpose. For example, create a group hive-users, and add the hive and impala users along with the users who will be loading data, to this group.

The example usermod and groupadd commands below are only applicable to locally defined groups on the NameNode, JobTracker, and ResourceManager. If you use another system for group management, equivalent changes should be made in your group management system.
$ groupadd hive-users
$ usermod -G someuser,hive-users someuser
$ usermod -G hive,hive-users hive

External Tables

External tables require the ALL@database privilege in addition to the URI privilege. When data is being inserted through the EXTERNAL TABLE statement, or is referenced from an HDFS location outside the normal Hive database directories, the user needs appropriate permissions on the URIs corresponding to those HDFS locations. This means that the URI location must either be owned by the hive:hive user OR the hive/impala users must be members of the group that owns the directory.

You can configure access to the directory using a URI as follows:
[roles]
someuser_home_dir_role = server=server1->uri=hdfs://ha-nn-uri/user/someuser
You should now be able to create an external table:
CREATE EXTERNAL TABLE ... 
LOCATION 'hdfs://ha-nn-uri/user/someuser/mytable';

Sample Sentry Configuration Files

This section provides a sample configuration.

Policy Files

The following is an example of a policy file with a per-DB policy file. In this example, the first policy file, sentry-provider.ini would exist in HDFS; hdfs://ha-nn-uri/etc/sentry/sentry-provider.ini might be an appropriate location. The per-DB policy file is for the customer's database. It is located at hdfs://ha-nn-uri/etc/sentry/customers.ini.

sentry-provider.ini
[databases]
# Defines the location of the per DB policy file for the customers DB/schema 
customers = hdfs://ha-nn-uri/etc/sentry/customers.ini 

[groups]
# Assigns each Hadoop group to its set of roles 
manager = analyst_role, junior_analyst_role
analyst = analyst_role
jranalyst = junior_analyst_role
customers_admin = customers_admin_role
admin = admin_role 

[roles]
# The uris below define a define a landing skid which
# the user can use to import or export data from the system.
# Since the server runs as the user "hive" files in that directory
# must either have the group hive and read/write set or
# be world read/write.
analyst_role = server=server1->db=analyst1, \
    server=server1->db=jranalyst1->table=*->action=select    
    server=server1->uri=hdfs://ha-nn-uri/landing/analyst1
junior_analyst_role = server=server1->db=jranalyst1, \
    server=server1->uri=hdfs://ha-nn-uri/landing/jranalyst1 

# Implies everything on server1 -> customers. Privileges for
# customers can be defined in the global policy file even though 
# customers has its only policy file. Note that the Privileges from
# both the global policy file and the per-DB policy file
# are merged. There is no overriding.
customers_admin_role = server=server1->db=customers 

# Implies everything on server1.
admin_role = server=server1
customers.ini
[groups]
manager = customers_insert_role, customers_select_role
analyst = customers_select_role 

[roles]
customers_insert_role = server=server1->db=customers->table=*->action=insert
customers_select_role = server=server1->db=customers->table=*->action=select

Sentry Configuration File

The following is an example of a sentry-site.xml file.

sentry-site.xml

<configuration>
  <property>
    <name>hive.sentry.provider</name>
    <value>org.apache.sentry.provider.file.HadoopGroupResourceAuthorizationProvider</value>
  </property>

  <property>
    <name>hive.sentry.provider.resource</name>
    <value>/path/to/authz-provider.ini</value>
    <!-- 
       If the hdfs-site.xml points to HDFS, the path will be in HDFS;
       alternatively you could specify a full path, e.g.:
       hdfs://namenode:port/path/to/authz-provider.ini
       file:///path/to/authz-provider.ini
    -->
  </property>

  <property>
    <name>sentry.hive.server</name>
    <value>server1</value>
  </property>
</configuration>

Accessing Sentry-Secured Data Outside Hive/Impala

When Sentry is enabled, the hive user owns all data within the Hive warehouse. However, unlike traditional database systems the enterprise data hub allows for multiple engines to execute over the same dataset.

However, there are scenarios where fully vetted and reviewed jobs will also need to access the data stored in the Hive warehouse. A typical scenario would be a secured MapReduce transformation job that is executed automatically as an application user. In such cases it's important to know that the user executing this job will also have full access to the data in the Hive warehouse.

Scenario One: Authorizing Jobs

Problem

A reviewed, vetted, and automated job requires access to the Hive warehouse and cannot use Hive/Impala to access the data.

Solution

Create a group which contains hive, impala, and the user executing the automated job. For example, if the etl user is executing the automated job, you can create a group called hive-users which contains the hive, impala, and etl users.

The example usermod and groupadd commands below are only applicable to locally defined groups on the NameNode, JobTracker, and ResourceManager. If you use another system for group management, equivalent changes should be made in your group management system.
$ groupadd hive-users
$ usermod -G hive,impala,hive-users hive
$ usermod -G hive,impala,hive-users impala
$ usermod -G etl,hive-users etl
Once you have added users to the hive-users group, change directory permissions in the HDFS:
$ hadoop fs -chgrp -R hive:hive-users /user/hive/warehouse
$ hadoop fs -chmod -R 770 /user/hive/warehouse

Scenario Two: Authorizing Group Access to Databases

Problem

One group of users, grp1 should have full access to the database, db1, outside of Sentry. The database, db1 should not be accessible to any other groups, outside of Sentry. Sentry should be used for all other authorization needs.

Solution

Place the hive and impala users in grp1.
$ usermod -G hive,impala,grp1 hive
$ usermod -G hive,impala,grp1 impala
Then change group ownerships of all directories and files in db1 to grp1, and modify directory permissions in the HDFS. This example is only applicable to local groups on a single host.
$ hadoop fs -chgrp -R hive:grp1 /user/hive/warehouse/db1.db 
$ hadoop fs -chmod -R 770 /user/hive/warehouse/db1.db

Debugging Failed Sentry Authorization Requests

Sentry logs all facts that lead up to authorization decisions at the debug level. If you do not understand why Sentry is denying access, the best way to debug is to temporarily turn on debug logging:
  • In Cloudera Manager, add log4j.logger.org.apache.sentry=DEBUG to the logging settings for your service through the corresponding Logging Safety Valve field for the Impala, Hive Server 2, or Solr Server services.
  • On systems not managed by Cloudera Manager, add log4j.logger.org.apache.sentry=DEBUG to the log4j.properties file on each host in the cluster, in the appropriate configuration directory for each service.
Specifically, look for exceptions and messages such as:
FilePermission server..., RequestPermission server...., result [true|false]
which indicate each evaluation Sentry makes. The FilePermission is from the policy file, while RequestPermission is the privilege required for the query. A RequestPermission will iterate over all appropriate FilePermission settings until a match is found. If no matching privilege is found, Sentry returns false indicating "Access Denied" .