[
https://issues.apache.org/jira/browse/HADOOP-14350?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16014717#comment-16014717
]
Steve Loughran commented on HADOOP-14350:
-----------------------------------------
OK, I understand.
# spark should be setting the absolute path
# hadoop should handle relative paths anyway.
Actually, Hadoop should have a version of the login command which takes a
{{File}} object and so remove the ambiguity. But that'd take a while to get
code downstream to switch to, even if we do it in Hadoop itself.
If you attach the spark stack, we can create a JIRA there and link it, so I'll
see about fixing it there while we do some absolution in Hadoop. Spark's
releasing faster, so this could maybe get to a 2.2.x release before long
> Relative path for Kerberos keytab is not working on IBM JDK
> -----------------------------------------------------------
>
> Key: HADOOP-14350
> URL: https://issues.apache.org/jira/browse/HADOOP-14350
> Project: Hadoop Common
> Issue Type: Bug
> Components: common, security
> Affects Versions: 2.7.3
> Environment: IBM JDK
> Reporter: Wen Yuan Chen
>
> For the sample code below:
> {code}
> public class TestKrb {
> public static void main(String[] args) throws IOException {
> String user = args[0], path = args[1];
> UserGroupInformation ugi =
> UserGroupInformation.loginUserFromKeytabAndReturnUGI(user, path);
> System.out.println("Login successfully");
> }
> }
> {code}
> When I use IBM JDK and pass a relative path for the Kerberos keytab, it will
> throw error messages. According to the debug log, it always tries to read
> the keytab from the root path. See the debug logs below:
> In above log, the useKeytab=<value> entry is showing a <value> prefaced by a
> leading "//". It appears that HADOOP is adjusting the user supplied keytab
> file and most likely prefacing it with something like "FILE://", which would
> cause the resulting IBM normalized value to then be prefaced by "//" before
> the user supplied keytab file. This is the cause for why relative paths used
> with HADOOP are not working with IBM JVM's.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
