Thanks Larry. This also seems to affect YARN as well when looking at the
YARN UI support in Apache Knox. Same sort of issue: "Logged in as: knox".

Kevin Risden

On Thu, May 3, 2018 at 10:10 AM, larry mccay <[email protected]> wrote:

> This can only be addressed in Hadoop, AFAICT.
> There are so many UIs and even APIs not supporting trusted proxies and it
> is really becoming a problem.
>
> We need to file JIRAs where this support is missing and potentially
> provide patches as it seems folks are reluctant to add proper support for
> it anymore.
>
> On Wed, May 2, 2018 at 2:56 PM, Kevin Risden <[email protected]> wrote:
>
>> There has been quite a bit of work done around enabling Knox and the
>> MapReduce Job History server UI. However, I think we are going to run into
>> a fundamental problem with the MapReduce Job History server and
>> impersonation/doAs.
>>
>> Based on our research, we found that the MR Job History server doesn't
>> support the "doAs" pattern that is used by Knox and Hue.
>>
>> If I navigate to a topology /jobhistory/ in the top right hand corner is:
>> "Logged in as: knox". Knox sends a doAs parameter (just like Hue) and it
>> seems like MR Job History server is ignoring this. Ideally the "Logged
>> in as:" would be the user who authenticated to Knox and not the user Knox
>> is running as (according to Kerberos).
>>
>> Right now this causes problems for us because the Knox user doesn't have
>> access to view the job logs. If I navigate to an attempt log it looks to
>> proxy correctly but end up with this error message:
>>
>> User [knox] is not authorized to view the logs for
>> attempt_1525122616004_0164_m_000000_0 in log file
>>
>> The same type of error happens in Hue when trying to look at job logs.
>>
>>
>> User [hue] is not authorized to view the logs for job_1521053483563_0133
>> in log file
>>
>>
>> It looks like this might have been previously reported as KNOX-747 but
>> was closed as can't reproduce.
>>
>> I don't think this is a problem with Knox but bringing it up here in case
>> there are any ideas on workarounds. This might have to be fixed upstream in
>> the Hadoop project.
>>
>> Any ideas on how to move forward with Knox and MapReduce Job History
>> server?
>>
>> Kevin Risden
>>
>
>

Reply via email to