[
https://issues.apache.org/jira/browse/HADOOP-5327?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12679539#action_12679539
]
Hemanth Yamijala commented on HADOOP-5327:
------------------------------------------
Few comments:
- There's an import of a LoginException in JobTracker, which is not required.
- I feel it would be better to overload checkAccess to have an additional
parameter UGI. This is because both modes (with and without UGI) are still
valid and hence it would be more convenient and fewer changes to code if we can
create the overloaded API.
- Can you please correct the comment on checkAccess, to remove references to
ownerAllowed, as it is no longer used. This is not introduced in this patch,
but since you are modifying the comment, it may make sense to correct this as
well.
- In the test cases, why is dfs.permissions set to false ? Is it because the
credentials of the new job (with invalid ACLs) is being submitted as a
different user and hence there will be problems to write to dfs ? Does this
happen, even though you are not using the MiniDFSCluster ?
- A general question is about whether any cleanup is required for the job which
fails ACLs. For instance, should we delete the job folder in the system
directory, delete the history files etc ? I am not sure if it is required. Any
thoughts ?
> Job files for a job failing because of ACLs are not clean from the system
> directory
> ------------------------------------------------------------------------------------
>
> Key: HADOOP-5327
> URL: https://issues.apache.org/jira/browse/HADOOP-5327
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Affects Versions: 0.20.0
> Reporter: Karam Singh
> Assignee: Amar Kamat
> Priority: Blocker
> Fix For: 0.20.0
>
> Attachments: HADOOP-5327-v2.3.patch
>
>
> Jobs which failed because of ACLs gets added during JT restart recovery
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
