hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Eric Sirianni (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HADOOP-7198) Hadoop defaults for web UI ports often fall smack in the middle of Linux ephemeral port range
Date Mon, 16 Dec 2013 18:49:09 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-7198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13849578#comment-13849578
] 

Eric Sirianni commented on HADOOP-7198:
---------------------------------------

+1

I've hit this several times now in our automated test environment when restarting the daemons.
 As Philip says, the default ports for most Hadoop daemons (e.g. 50010 for DataNode, etc.)
fall within Linux's ephemeral port range.  What's the downside to making the Hadoop default
ports less than 32768 to remove the possibility of conflict with ephemeral ports?

> Hadoop defaults for web UI ports often fall smack in the middle of Linux ephemeral port
range
> ---------------------------------------------------------------------------------------------
>
>                 Key: HADOOP-7198
>                 URL: https://issues.apache.org/jira/browse/HADOOP-7198
>             Project: Hadoop Common
>          Issue Type: Wish
>            Reporter: Philip Zeyliger
>            Priority: Trivial
>
> It turns out (see http://en.wikipedia.org/wiki/Ephemeral_port and  /proc/sys/net/ipv4/ip_local_port_range)
that when you bind to port 0, Linux chooses an ephemeral port.  On my default-ridden Ubuntu
Maverick box and on CentOS 5.5, that range is 32768-61000.  So, when HBase binds to 60030
or when mapReduce binds to 50070, there's a small chance that you'll conflict with, say, an
FTP session, or with some other Hadoop daemon that's had a listening address configured as
:0.
> I don't know that there's a practical resolution here, since changing the defaults seems
like an ill-fated effort, but if you have any ephemeral port use, you can run into this. 
We've now run into it once.



--
This message was sent by Atlassian JIRA
(v6.1.4#6159)

Mime
View raw message