[StarCluster] creating a new AMI for starcluster -- can't log in

[Dan Tenenbaum]

On Sat, Oct 30, 2010 at 8:54 AM, Justin Riley <jtriley at mit.edu> wrote:

>  On 10/29/10 8:07 PM, Dan Tenenbaum wrote:
> > Could this have something to do with the preparation that starcluster
> > does prior to creating an image?
>
> Well if you didn't set disable_root: 0 then not being able to login as
> root is most likely because of that. Not being able to login as ubuntu
> user though *could* be caused by the cloud-init having a new behavior in
> 10.10 that it does not setup /home/ubuntu/.ssh folder if it does not
> already exist (that is purely a guess, I haven't looked at the latest
> version of cloud-init since the 10.04 Alestic AMI). The preparation in
> starcluster 0.91.2's createimage command does blast all .ssh folders in
> /home before imaging so if the new cloud-init scripts do have this
> behavior then that would be why you can't login to ubuntu user after
> imaging with createimage.
>

I ended up following the directions on the wiki, and using 10.04.. I did
make the change setting disable_root to 0.

I had done the same thing on 10.10 but it didn't take for some reason.

Now I can ssh in to my instances as either root or sgeadmin, but when I
start a cluster, I cannot ssh from master to a worker as sgeadmin without 1)
being prompted about the key and 2) getting "permission denied (publickey)".
I can work around this by doing all my work as root, but it's not ideal.

According to the logs, sgeadmin's rsa key is being created:
clustersetup.py:144 - INFO - Configuring passwordless ssh for user: sgeadmin
ssh.py:251 - DEBUG - True
ssh.py:251 - DEBUG - True
clustersetup.py:158 - DEBUG - Checking for orphaned private key file:
/home/sgeadmin/.ssh/id_rsa | exists = True
clustersetup.py:160 - DEBUG - Removing orphaned private key file:
/home/sgeadmin/.ssh/id_rsa
clustersetup.py:158 - DEBUG - Checking for orphaned private key file:
/home/sgeadmin/.ssh/id_rsa.pub | exists = True
clustersetup.py:160 - DEBUG - Removing orphaned private key file:
/home/sgeadmin/.ssh/id_rsa.pub
clustersetup.py:162 - INFO - Generating local RSA ssh keys for user:
sgeadmin

id_rsa and id_rsa.pub DO exist in ~sgeadmin/.ssh on the master and workers.
So I"m not sure what's up with that.

Thanks
Dan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/pipermail/starcluster/attachments/20101030/b6d7a262/attachment.htm