[
https://issues.apache.org/jira/browse/HADOOP-6520?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12806266#action_12806266
]
Devaraj Das commented on HADOOP-6520:
-------------------------------------
This environment variable exists - JOB_TOKEN_FILE. In this jira, we should read
the contents of the file via TokenCache's APIs and load the tokens in the UGI
(similar to the way job-token is loaded in Child.java). In the JobInProgress,
the tokens already exist in memory and it should be easy to add it to the ugi
used for copying the jobconf file from the hdfs. In the tasktracker, ditto (for
localizing the job's files).
For tasks (and for other processes like an MR Streaming app running a hdfs
command from within the tasks), if we have a static block in the UGI to look
for the env variable, and load the token cache using the TokenCache's APIs, it
should work. Also, when the env var is defined, we should use simple
authentication for login.
Thoughts?
> UGI should load tokens from the environment
> -------------------------------------------
>
> Key: HADOOP-6520
> URL: https://issues.apache.org/jira/browse/HADOOP-6520
> Project: Hadoop Common
> Issue Type: Bug
> Reporter: Owen O'Malley
>
> For MapReduce tasks, we need not only the task itself, but child processes to
> get the delegation tokens, therefore we need an environment variable that if
> set loads a token cache into the login UGI. The task tracker can set this
> property so that tasks (and child processes of the tasks, such as streaming
> and pipes) have the job's tokens automatically.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
