Mailing-List: contact common-issues-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: common-issues@hadoop.apache.org
Date: Tue, 31 Mar 2015 21:22:56 +0000 (UTC)
From: "Allen Wittenauer (JIRA)" <jira@apache.org>
To: common-issues@hadoop.apache.org
Message-ID: <JIRA.12780957.1426015870000.86240.1427836976051@Atlassian.JIRA>
In-Reply-To: <JIRA.12780957.1426015870000@Atlassian.JIRA>
References: <JIRA.12780957.1426015870000@Atlassian.JIRA>
 <JIRA.12780957.1426015870004@arcas>
Subject: [jira] [Comment Edited] (HADOOP-11764) Hadoop should have the
 option to use directory other than tmp for extracting and loading
 leveldbjni
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/HADOOP-11764?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14389438#comment-14389438 ] 

Allen Wittenauer edited comment on HADOOP-11764 at 3/31/15 9:22 PM:
--------------------------------------------------------------------

bq. However, I agree with Anubhav Dhoot that it's better to setup env vars instead of modifying each daemon to set system property programatically.

... except, you don't have to.  All of the daemons run through the Configuration class (and probably other classes as well).  It could get set then, *if it isn't already*.

bq. Would you mind elaborating what is the complete mess? The admin just needs to set HADOOP_LEVELDBJNI_DIR to somewhere else than /tmp if /tmp is mounted as noexec. If HADOOP_LEVELDBJNI_DIR is not set, it will just work as what it is now.

I'm specifically referring to:

bq. Therefore, my proposal is if HADOOP_LEVELDBJNI_DIR is specified, it will set to library.leveldbjni.path as well as be added to JAVA_LIBRARY_PATH.

So not only do we have an extra -D parameter, -Djava.library.path also has the exact same content which increases the line length even more.  This will almost certainly break some users at some point in time.


was (Author: aw):
bq. However, I agree with Anubhav Dhoot that it's better to setup env vars instead of modifying each daemon to set system property programatically.

... except, you don't have to.  All of the daemons run through the Configuration class.  It could get set then, *if it isn't already*.

bq. Would you mind elaborating what is the complete mess? The admin just needs to set HADOOP_LEVELDBJNI_DIR to somewhere else than /tmp if /tmp is mounted as noexec. If HADOOP_LEVELDBJNI_DIR is not set, it will just work as what it is now.

I'm specifically referring to:

bq. Therefore, my proposal is if HADOOP_LEVELDBJNI_DIR is specified, it will set to library.leveldbjni.path as well as be added to JAVA_LIBRARY_PATH.

So not only do we have an extra -D parameter, -Djava.library.path also has the exact same content which increases the line length even more.  This will almost certainly break some users at some point in time.

> Hadoop should have the option to use directory other than tmp for extracting and loading leveldbjni
> ---------------------------------------------------------------------------------------------------
>
>                 Key: HADOOP-11764
>                 URL: https://issues.apache.org/jira/browse/HADOOP-11764
>             Project: Hadoop Common
>          Issue Type: Bug
>            Reporter: Anubhav Dhoot
>            Assignee: Anubhav Dhoot
>         Attachments: YARN-3331.001.patch, YARN-3331.002.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
(v6.3.4#6332)