[Hadoop-common-user] openssh - can't achieve passphraseless ssh

More then likely this is the permissions of the authorized keys file.
Make sure:
chmod 600 ~/.ssh/authorized_keys
Make sure:
file is owned by proper user
Make sure:
drwx------ .ssh
You can tune up the verbosity of your ssh server to troubleshoot this more.

There are lots of ssh key tutorials out there. Good hunting.

Another sneaky permissions requirement is that ~/.ssh/ itself must be mode
0750 or more strict.

- Aaron

Success! Thank you Edward and Aaron. I had to change the permissions of
my home directory "chmod 750 ~" and chown my home dir so that I owned
it.

FYI: when debugging sshd, the -d and -e options are very useful.

So I ran the 0.20.1 grep example is pseudo distributed mode and it
worked. It was much slower than the run I did in Standalone mode. I'm
running Windows XP and the latest version of cygwin, 1.5.25-15. Is
Windows generally slow or is it a tuning issue?

Thanks again.

-Dennis

-----Original Message-----
From: Aaron Kimball
Sent: Thursday, October 22, 2009 12:28 AM
To: [email protected]
Subject: Re: openssh - can't achieve passphraseless ssh

Another sneaky permissions requirement is that ~/.ssh/ itself must be
mode
0750 or more strict.

- Aaron

On Wed, Oct 21, 2009 at 2:47 PM, Edward Capriolo
wrote:
1.5.25-15
and I

Hi Dennis,
This is normal to see pseudodistributed much slower than standalone.

Most of the extra overhead you're seeing is due to the heartbeat mechanism
of Hadoop. When the JobTracker has tasks that need to be run, it does not
actually push them to the TaskTrackers. Instead, the TaskTrackers
periodically heartbeat to the JobTracker asking for new tasks. This is true
even on a single node pseudodistributed cluster. Because of this, there is a
fair amount of wall clock time spent sleeping waiting for heartbeats to
occur so the tasks can get assigned.

This might sound like a big problem, but Hadoop is just not designed for
"interactive" jobs that only take a few seconds. At last count, I found that
"no-op" jobs that only process a few bytes of data usually take between 20
and 25 seconds on a pseudodistributed cluster. The problem is a bit less
pronounced on a larger cluster since tasktrackers are heartbeating more
regularly.

-Todd