Physical Machines

Openstack depends on gomes and our R900s, plus to a lesser extent (via ceph) the other management nodes and the sunfires. Our ceph infrasturcture is covered in greater depth :doc:elsewhere <../storage/ceph>; this document is concerned with what you need to know to run the Openstack-specific machines.

Openstack Management on Gomes

Processes

Gomes should be running a variety of openstack processes. This list, grouped by openstack component, will tell you what they are (called, at least):

  • Nova processes
  • Keystone processes
  • Neutron processes
  • Cinder processes
  • Glance processes

Homegrown Scripts

usered
A “simple” perl script for messing around with keystone, created to tide us over until openstack has a real command-line client. Lives in the usual spot in afs, under admins.pub/scripts.
fuck...directly...
Pretty much exactly what it says on the tin. Sometimes you just need to fuck...directly.... Lives in localadmin’s homedir on gomes. Currently mildly broken.

Futzing with the Database

See these articles for more, but:

There are some things that the Openstack commands can’t do, like recover nicely from failures. In those cases, you should (well…) manually modify the datbase on gomes (run sudo su postgres -c psql), to manually modify the database.

The Compute Nodes

Antonio, Enrique, and Serrao are our three current R900s. They are (at least in theory) identically installed, so the following applies to all of them.

When a Compute Node Goes Down

Follow the instructions here. You’ll need to modify the database (per those instructions) to migrate vms over. Note that, deleted vms will still be in the instances table, just with deleted set to 1, and that you should keep a list of the vms you’re migrating. For each migrated vm, you’ll need to make an empty directory (owned by nova:nova) on the new compute node called /var/lib/nova/instances/MIGRATED-VM-UUID.