Cross Kickstarting in Rocks 4.3

So you want to cross Kickstart nodes that aren't the same architecture as your front end? Don't worry, rocks can do that, or it's supposed to be able to. Either way, here's how you'd do it. But first somethings you'll need.

You'll need:

* A Kernel/Boot CD in the Native architecture of your Node.

* All the required Rolls of your Nodes architecture (which are listed at the download location).

What to do:

Get the rolls from rocks in the architecture of your node, you can find them here.

Now you've got to add the rolls to your local mirror.

If you downloaded the .iso, don't bother burning it to CD use this command:

# mount -o loop <roll-name>.iso /mnt/cdrom

Then add the rolls from the CD into your local mirror with:

# rocks-dist --install copyroll

Unmount that .iso and then rebuild your distribution with the following flags, after you've built your distribution for your native architecture first.

For an x86_64 Node:

# cd /home/install
# rocks-dist --arch=x86_64 dist

For an i386 Node:

# cd /home/install
# rocks-dist --arch=i386 dist

Now, just boot your Node from a Kernel/Boot CD of it's native architecture and install, it should be caught by insert-ethers on your front end (which you'll have to run). Be aware Rocks does not support PXE cross-kickstart installs, though some people have gotten it working themselves. You'll have to install from a native boot disk.

Once installed, you will be able to Kickstart those nodes, so from reading through the mailing list archives, you should only have to "touch" each node once for just the install, from then on you should be able to kickstart those nodes from your front end.

Our troubles with getting this running:

Right now, we can't seem to get insert-ethers on umopt1 to notice an installing node. Bob has expressed the concern that this might not work for a frontend in x86_64 to cross-kickstart i386 nodes, since the 4.3 documentation doesn't have any examples of it, I'm hoping that they were just being unreasonably terse, as the later documentation (5.1) does contain examples of our scenario.

Fix: Alright so when rocks is running insert-ethers it's looking for failed DCHP requests, which are generated by the syslog service, well sometimes that prevents Rocks from reading them, or gets in the way, so a simple:

# service syslog restart

Fixed the whole problem.

Right now however, we don't have a set of rpms to build the new nodes, but they can begin installing from the headnode, so once we have a running batch of rpms, we'll be good to go.

-- JamesWright - 16 Jun 2009