So now that we can script stateless provisioning, we might also want to use stateful provisioning. That is PXE boot node once to format drive and install kernel + OS but then disable PXE boot and boot from hard drive. Use scenario, I have some nodes with large fast disks on board which can provide local scratch area.
Documentation page (http://warewulf.lbl.gov/trac/wiki/Recipes/StatefulProvisioning)
From user list (Jason Stover) with modified notes:
1) Install Kernel and Grub packages into VNFS
From the list “The main concern I can think of here is in the image (VNFS+files) size. If you have a large, non-hybridized configuration you use for statefull deployment, you may want to setup hybridization, or in general trim your image down to run statelessly due to potential memory usage concerns. Other than that, using a single image for both stateless/statefull will work just fine.”
So we'll stick with one chroot for now until it becomes a problem. So we can go back and forth with bootlocal option and keep the stateless vnfs image for PXE boot in database.
2) Be sure you're configuring BOOTLOADER, DISKPARTITION, DISKFORMAT, and FILESYSTEMS
3) Have the node PXE boot. Once it has provisioned, set bootlocal to 1:
4) Reboot node. Still with PXE as the first boot device. The PXE config file causes it to boot off hard drive instead of image if bootlocal is set. So, it “should” be booting of the HDD.
Update: Finally got back to this project. Just making notes here as I'm not going to pursue this. Building up a VNFS is tedious. I'd rather build up a server and then grab a
golden image. So I'm going to pursue that angle. Stateless provisioning would be useful for me in the future with many node deployments of simple compute servers. Warewulf Stateless
— Meij, Henk 2015/12/04 14:00