[
https://issues.apache.org/jira/browse/HADOOP-16023?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16813616#comment-16813616
]
Eric Yang commented on HADOOP-16023:
------------------------------------
[~daryn] When RM's hadoop auth_to_local, maps johnSmith to johnsmith. Node
manager receives the UGI call, and starts container-executor using johnsmith
[username
string|https://github.com/apache/hadoop/blob/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/impl/main.c#L673].
User johnSmith's program reads ~/.ssh/id_rsa key, and print it to log. This
is the default behavior when using LinuxContainerExecutor. There is no setting
required.
> Support system /etc/krb5.conf for auth_to_local rules
> -----------------------------------------------------
>
> Key: HADOOP-16023
> URL: https://issues.apache.org/jira/browse/HADOOP-16023
> Project: Hadoop Common
> Issue Type: Improvement
> Reporter: Bolke de Bruin
> Assignee: Bolke de Bruin
> Priority: Major
> Labels: security
>
> Hadoop has long maintained its own configuration for Kerberos' auth_to_local
> rules. To the user this is counter intuitive and increases the complexity of
> maintaining a secure system as the normal way of configuring these
> auth_to_local rules is done in the site wide krb5.conf usually /etc/krb5.conf.
> With HADOOP-15996 there is now support for configuring how Hadoop should
> evaluate auth_to_local rules. A "system" mechanism should be added.
> It should be investigated how to properly parse krb5.conf. JDK seems to be
> lacking as it is unable to obtain auth_to_local rules due to a bug in its
> parser. Apache Kerby has an implementation that could be used. A native (C)
> version is also a possibility.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
