Provisioning

Overview

The process of provisioning nodes in a TryStack zone is an evolving practice. Here we describe what we've done for the first and second zones, and the model we'd like to move towards.

Provisioning Process for Zone A

For the first zone -- Zone A - the DELL zone -- Nati Ueno adapted the build_pxe_env.sh script from the Devstack project to create a PXE install server on the management node. The PXE install environment publishes the base operating system image (Ubuntu 11.04 for Zone A) that all other service nodes netboot into when kicked.

The base OS image that is used on the service nodes is a slightly modified Ubuntu Natty 11.04 server image. The modifications made to the stock netboot image are:

• Installation of the kvm, libvirt-bin, kpartx, cloud-utils and curl packages
• Installation of the chef-client package, the Chef/Opscode keyring management and configuring it to point to the management node's chef-server
• Installation of the mdadm RAID administration package
• Copying /etc/sudoers and /etc/rc.local files that each service node shares
• Copying of the management node's /etc/resolv.conf file
• Creation of a standard /etc/networks/interfaces file that brings up eth0 via DHCP

Once the base operating system image is installed into the PXE install environment, we can kick a service node via IPMI to cause it to hard reset and netboot into the base image. Once booted, the rc.local script creates the RAID array, mounts the single partition to /opt/openstack, and symbolically links /etc/chef with /opt/openstack/chef.

Provisioning Process for Zone B

We're trying out a different provisioning process in Zone B -- the HP zone. On Zone B's management node, we are installing Cobbler and using Cobbler's functionality to establish a PXE install server and manage service node information.