The groups of a user is determined by a group mapping service provider.
Hadoop supports various group mapping mechanisms, configured by the `hadoop.security.group.mapping` property. Some of them, such as `JniBasedUnixGroupsMappingWithFallback`, use operating
systems' group name resolution and requires no configuration. But Hadoop also supports special group mapping mechanisms through
LDAP and composition of LDAP and operating system group name resolution, which require additional configurations.
`hadoop.security.group.mapping` can be one of the following:
The default implementation. It will determine if the Java Native Interface (JNI) is available. If JNI is available, the implementation will use the API within hadoop to resolve a list of groups for a user. If JNI is not available then the shell-based implementation, `ShellBasedUnixGroupsMapping`, is used.
Similar to `JniBasedUnixGroupsMappingWithFallback`. If JNI is available, it obtains netgroup membership using the Hadoop native API; otherwise uses `ShellBasedUnixGroupsNetgroupMapping`.
This implementation shells out with the `bash -c groups` command (for a Linux/Unix environment) or the `net group` command (for a Windows environment) to resolve a list of groups for a user.
This implementation is similar to `ShellBasedUnixGroupsMapping`, except that it executes `getent netgroup` command to get netgroup membership.
***org.apache.hadoop.security.LdapGroupsMapping**
An alternate implementation, which connects directly to an LDAP server to resolve the list of groups. However, this provider should only be used if the required groups reside exclusively in LDAP, and are not materialized on the Unix servers.
LdapGroupsMapping supports SSL connection and POSIX group semantics. See section [LDAP Groups Mapping](#LDAP_Groups_Mapping) for details.
This implementation composites other group mapping providers for determining group membership. This allows to combine existing provider implementations and composite a virtually new provider without customized development to deal with complex situation. See section [Composite Groups Mapping](#Composite_Groups_Mapping) for details.
For HDFS, the mapping of users to groups is performed on the NameNode. Thus, the host system configuration of the NameNode determines the group mappings for the users.
Note that HDFS stores the user and group of a file or directory as strings; there is no conversion from user and group identity numbers as is conventional in Unix.
It is possible to statically map users to groups by defining the mapping in `hadoop.user.group.static.mapping.overrides` in the format `user1=group1,group2;user2=;user3=group2`.
This property overrides any group mapping service provider. If a user's groups are defined in it, the groups are returned without more lookups; otherwise, the service provider defined in `hadoop.security.group.mapping` is used to look up the groups. By default, `dr.who=;` is defined, so the fake user dr.who will not have any groups.
Caching/Negative caching
--------
Since the group mapping resolution relies on external mechanisms, the NameNode performance may be impacted. To reduce the impact due to repeated lookups, Hadoop caches the groups returned by the service provider. The cache invalidate is configurable via `hadoop.security.groups.cache.secs`, and the default is 300 seconds.
To avoid spamming NameNode with unknown users, Hadoop employs negative caching so that if the result of the lookup is empty, return an empty group directly instead of performing more group mapping queries,
The cache invalidation is configurable via `hadoop.security.groups.negative-cache.secs`. The default is 30 seconds, so if group mapping service providers returns no group for a user, no lookup will be performed for the same user within 30 seconds.
This provider supports LDAP with simple password authentication using JNDI API.
`hadoop.security.group.mapping.ldap.url` must be set. This refers to the URL of the LDAP server for resolving user groups.
`hadoop.security.group.mapping.ldap.base` configures the search base for the LDAP connection. This is a distinguished name, and will typically be the root of the LDAP directory.
If the LDAP server does not support anonymous binds,
set the distinguished name of the user to bind in `hadoop.security.group.mapping.ldap.bind.user`.
The path to the file containing the bind user's password is specified in `hadoop.security.group.mapping.ldap.bind.password.file`.
This file should be readable only by the Unix user running the daemons.
It is possible to set a maximum time limit when searching and awaiting a result.
Set `hadoop.security.group.mapping.ldap.directory.search.timeout` to 0 if infinite wait period is desired. Default is 10,000 milliseconds (10 seconds).
The implementation does not attempt to resolve group hierarchies. Therefore, a user must be an explicit member of a group object
in order to be considered a member.
### Active Directory ###
The default configuration supports LDAP group name resolution with an Active Directory server.
To secure the connection, the implementation supports LDAP over SSL (LDAPS). SSL is enable by setting `hadoop.security.group.mapping.ldap.ssl` to `true`.
In addition, specify the path to the keystore file for SSL connection in `hadoop.security.group.mapping.ldap.ssl.keystore` and keystore password in `hadoop.security.group.mapping.ldap.ssl.keystore.password`.
Alternatively, store the keystore password in a file, and point `hadoop.security.group.mapping.ldap.ssl.keystore.password.file` to that file. For security purposes, this file should be readable only by the Unix user running the daemons.
Typically, Hadoop resolves a user's group names by making two LDAP queries: the first query gets the user object, and the second query uses the user's Distinguished Name to find the groups.
For some LDAP servers, such as Active Directory, the user object returned in the first query also contains the DN of the user's groups in its `memberOf` attribute, and the name of a group is its Relative Distinguished Name.
Therefore, it is possible to infer the user's groups from the first query without sending the second one, and it may reduce group name resolution latency incurred by the second query. If it fails to get group names, it will fall back to the typical two-query scenario and send the second query to get group names.
To enable this feature, set `hadoop.security.group.mapping.ldap.search.attr.memberof` to `memberOf`, and Hadoop will resolve group names using this attribute in the user object.
`CompositeGroupsMapping` works by enumerating a list of service providers in `hadoop.security.group.mapping.providers`.
It get groups from each of the providers in the list one after the other. If `hadoop.security.group.mapping.providers.combined` is `true`, merge the groups returned by all providers; otherwise, return the groups in the first successful provider.
See the following section for a sample configuration.
### Multiple group mapping providers configuration sample ###
This sample illustrates a typical use case for `CompositeGroupsMapping` where
Hadoop authentication uses MIT Kerberos which trusts an AD realm. In this case, service
principals such as hdfs, mapred, hbase, hive, oozie and etc can be placed in MIT Kerberos,
but end users are just from the trusted AD. For the service principals, `ShellBasedUnixGroupsMapping`
provider can be used to query their groups for efficiency, and for end users, `LdapGroupsMapping`
provider can be used. This avoids to add group entries in AD for service principals when only using
`LdapGroupsMapping` provider.
In case multiple ADs are involved and trusted by the MIT Kerberos, `LdapGroupsMapping`
provider can be used multiple times with different AD specific configurations. This sample also shows how
to do that. Here are the necessary configurations.
