Allen Wittenauer commented on YARN-3331:

We really can't read this in via the core-site.xml file?  I could have sworn we 
populated the system properties with the content of the xml files on startup.  
It's really less than ideal to set this up on the command line as it is already 
quite long and there is a risk that we'll overflow the buffer, preventing 
things from running.

That issue aside:

a) We need to avoid using daemon-specific and/or project-specific environment 
variables.  This is the #1 reason why the branch-2 shell code is just utter 
chaos.  This should be HADOOP_ something.

b) Additionally, this should be set for everything, not just the nodemanager in 
order to keep this consistent if/when other stuff uses this functionality. The 
add_param should probably happen in finalize_hadoop_opts

c) The default should be set with an actual value in hadoop_basic_init.  $(pwd) 
is a *terrible* default value for this, because there is no guarantee what the 
pwd actually is.

d) There needs to be a stanza added to hadoop-env.sh that actually explains 
what your new environment variable does, how to set it, etc.

All/most of this is covered on 
http://wiki.apache.org/hadoop/UnixShellScriptProgrammingGuide .  I'm now going 
to go check to make sure I didn't miss anything on that page and add it if I 
did. :)

> NodeManager should use directory other than tmp for extracting and loading 
> leveldbjni
> -------------------------------------------------------------------------------------
>                 Key: YARN-3331
>                 URL: https://issues.apache.org/jira/browse/YARN-3331
>             Project: Hadoop YARN
>          Issue Type: Bug
>          Components: nodemanager
>            Reporter: Anubhav Dhoot
>            Assignee: Anubhav Dhoot
>         Attachments: YARN-3331.001.patch
> /tmp can be  required to be noexec in many environments. This causes a 
> problem when  nodemanager tries to load the leveldbjni library which can get 
> unpacked and executed from /tmp.

This message was sent by Atlassian JIRA

Reply via email to