Ubuntu – Using Juju returns “ssh invalid key” error

The solution I came up with is to set a password for the newly-booted nodes, and then manually insert SSH keys into each of them. To set a boot password for the ubuntu user, ensure that the following lines to /var/lib/cobbler/kickstarts/maas.preseed:

d-i    passwd/make-user boolean true
d-i    passwd/user-fullname ubuntu
d-i    passwd/username string ubuntu
d-i    passwd/user-password-crypted password <CRYPTED PASSWORD>

Once this is done, you can ssh ubuntu@ and use the password specified in the crypted password string (easiest way is to use one from an /etc/shadow file you already know) to log in. You can then insert your SSH public keys under ~ubuntu/.ssh/authorized_keys and ~root/.ssh/authorized keys.

Note that this is a workaround- once you ssh-keygen, MaaS should either be pulling the id_rsa.pub from your .ssh directory, or from the MaaS WebUI, where the user can specify a public key in his profile. No matter what I've tried, these keys aren't propagated, so I came up with the workaround.

A further cheat is to just add the .pub key to your MaaS node's .ssh/authorized_keys and then scp it to each of the nodes in the MaaS:

for i in `cobbler system list |grep -v default`; 
    do j=`cobbler system dumpvars --name "${i}" | grep hostname |grep -v duplicate |cut -f 2 -d \:`; 
    scp ~/.ssh/authorized_keys ubuntu@${j}:.ssh/authorized_keys;
done

This leaves you able to just repeatedly accept the SSH certificate errors and type the password in the crypted string to populate your entire MaaS with the SSH public key.

You should leave PXE booting on at all times. MAAS will tell the machine to local boot as appropriate depending on what state it is in.

I don't know why your agent isn't restarting between reboots, it sounds like a juju bug but I am not a juju expert (I work on MAAS). After you do a juju bootstrap, the node installs itself and you'll see it reboot at the end of that. You should see messages going to the console saying the agent is starting up. That reboot is no different to you rebooting it manually later so I suspect that something went wrong during installation and the installer didn't manage to tell MAAS to turn off the "netboot" flag. Can you see any obvious errors on the console when it installs the node during deployment?