[torqueusers] Passwordless_ssh on a diskless cluster

Daniel Bourque dbourque at weatherdata.com
Mon Apr 28 16:34:36 MDT 2008


first, if you remotely did a "/etc/init.d/sshd restart" , well you 
kicked yourself out. restart run stop then start, connection drops after 
stop. Always do "nohup /etc/init.d/sshd restart"

since you're using diskless, go on the nfs homedir, run

ssh-keygen -t dsa ( or rsa )
cd .ssh
cat id_dsa.pub > authorized_keys
chmod 600 authorized_keys

I'm assuming that the headnode and compute node shares the same nfs 
mounted home dir, and that you want to allow passwordless logins in 
either directions.

Hope this helps

Daniel Bourque
Sr. Systems Engineer
WeatherData Service Inc
An Accuweather Company



Gelonia L Dent wrote:

>Dear All,
>
>I am trying to eliminate using password login between the headnode and
>compute nodes on a diskless cluster. I've using the following steps:
>
>____________________
>On head:
>ssh-keygen -t rsa         (didn't enter a pass phrase)
>ssh-copy-id -i ~/.ssh/id_rsa.pub nodeN   (for a subset of the nodes
>(N=0,126))
>
>On compute nodes
>ssh-keygen -t rsa
>ssh-copy-id -i ~/.ssh/id_rsa.pub master
>______________________
>
>I have three issues:
>First, I did this on ten nodes, then something happened and I had to
>restart the sshd daemon, now I've lost connection to those ten nodes
>completely. I think my authorized_keys file got overwritten when I
>restarted ssh. How can I reach these nodes again?
>      Error message/       ssh: connect to host node10 port 22: No route
>to host
>
>Second, the first time I attempted this (before I broke those 10 nodes) I
>neglected to copy the same id_rsa.pub file to all the nodes and was
>generating a different one for each node. I've since corrected that step
>and am copying the same *.pub file from the head to all the nodes, then
>generating a *.pub from each node to the head. This gives me
>bi-directional access between each node and the head, but not between
>nodes. Do I need passwordless access between nodes for Moab/Torque to work
>properly? If so, how can this be accomplished.
>
>Third, today I am attempting again to go through this process. I have 128
>nodes. Is there a parameter in the sshd_config  file that is set to use
>passwords across the nodes which I may have overlooked. If so, can it be
>used to set a 'no password option' permanently between the head and
>compute nodes.
>
>Any ideas would be greatly appreciated.
>
>Best,
>
>--
>Gelonia Dent, PhD
>Manager of Scientific Computing
>Invertebrate Zoology
>The American Museum of Natural History
>(212) 313-7911
>
>
>
>_______________________________________________
>torqueusers mailing list
>torqueusers at supercluster.org
>http://www.supercluster.org/mailman/listinfo/torqueusers
>
>  
>


More information about the torqueusers mailing list