[
https://issues.apache.org/jira/browse/YARN-7430?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16244443#comment-16244443
]
Eric Yang commented on YARN-7430:
---------------------------------
[~shaneku...@gmail.com] . Thank you for explaining your point of view. I
understand how you arrived at these conclusions, but some use cases can not be
satisfied by the current implementation.
{quote}
User "foo" in the container does not have permission to execute the launch
script owned by "skumpf" and thus the container will fail to launch with a
permission denied error. We need the -user/uid option even if privileged is
requested, because without it, we have no idea what user the container will run
as.
{quote}
What is the point of using privileged flag, if the process can only run as
"skump" to run properly for privileged container? When container is granted
with root power, root user should have ability to do anything, why drop that
privilege away then reacquire it later using sticky bit? It is counter
intuitive.
Let review the ground rules that docker recommends, and what we are
recommending to Hadoop users.
# Docker security document clearly stated that docker must be run by trusted
user only. This means user either have sudo privileges or they are part of
docker group.
# Privileged container allows ENTRYPOINT to spawn multi-user environment such
as systemd or init like environment for multi-user support.
# Hadoop YARN user can be a trusted user to spawn docker containers on behave
of the end user.
# Hadoop simulates doAs call through container-executor, therefore docker
security recommendation stay intact. If container must run for end user who
isn't part of privileged user nor docker group, then precaution must be taken
to secure point of entry by yarn user or container-executor.
# Docker does not know about external users and group on LDAP, hence use of
{{--user [username]}} is essentially limited to container's {{/etc/passwd}} and
{{/etc/group}} to lookup group membership. Users/Group can be programmed into
docker container build, however this solution can not be generalized for LDAP
users in Hadoop eco-system. We don't want to end up rebuilding images, each
time a new LDAP user is added.
# Docker added {{--user uid:gid}} and {{--group-add}} to assign user credential
and group membership without the user depends on /etc/passwd and /etc/group for
lookup for dynamic users.
In order to resolve the conflicting user management between docker and Hadoop.
We must streamline the implementation to have capacity of supporting
multi-users docker container (privileged container) as well as single LDAP user
container (non-privileged container). Privileged container can only be spawned
by trusted user for trusted user. Hence, the privileged container image can
contain multiple users that is already pre-approved by system administrator.
Privileged container can acquire additional resources using mount points, and
consistent file system ACL inside and outside of container governs the overall
security.
There should never be a case where we allow localized resource for {{skump}} to
work as {{foo}} user without properly secure file system ACL. At least we
don't want to make this case work to ensure file system ACL rules are not
broken. Ultimately, file system permission is the last line of security
defense that we have for storing files in HDFS via NFS mount point.
>From this point of view, does it make more sense to run {{--privileged}}
>without {{--user username}}?
> User and Group mapping are incorrect in docker container
> --------------------------------------------------------
>
> Key: YARN-7430
> URL: https://issues.apache.org/jira/browse/YARN-7430
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: security, yarn
> Affects Versions: 2.9.0, 3.0.0
> Reporter: Eric Yang
> Assignee: Eric Yang
> Priority: Blocker
> Attachments: YARN-7430.001.patch
>
>
> In YARN-4266, the recommendation was to use -u [uid]:[gid] numeric values to
> enforce user and group for the running user. In YARN-6623, this translated
> to --user=test --group-add=group1. The code no longer enforce group
> correctly for launched process.
> In addition, the implementation in YARN-6623 requires the user and group
> information to exist in container to translate username and group to uid/gid.
> For users on LDAP, there is no good way to populate container with user and
> group information.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
