Ceph » devops

A few possible kinks:

• The fabfile currently pulls the magic ssh keys for rsync from cephbooter. we need a new location for those keys that lives on the secure network, or something equivalent.
• The ceph.com 'custom' param on the uebernet has the /gitbuilder-foo/ ProxyPass lines going to gitbuilder-foo.ceph.newdream.net. those should be switched to gitbuilder-foo.dynamicspecialness.sepia.ceph.com when the new vms come up.
• The only futzing I had to do with the old vms from base precise was to add an ssh key for fabric to get in and to add NOPASSWD to /etc/sudoers for.. group sudoers? I forget. (By default the passwordless sudo would work from the console but not over ssh, or something similarly annoying.)
• I can't remember if the debian images were fully tested from scratch. Those might need some futzing (creating debian user, etc.).

for the record, this is being done in this sprint for two reasons:
(a) we are under some pressure to give up the old gitbuilder hardware/rack-space.
(b) moving the gitbuilder will bump into a few problems in its current
implementation, necessitating some cleanup ... so we'll take a few days
and try to make this basic tool much better.

I moved http://gitbuilder-precise-i386.front.sepia.ceph.com/ and wrote a doc on how to move whole images: https://docs.google.com/a/inktank.com/document/d/1HJAEgcN-fsUCgHTjuNmG3NiI-MtQhBjD8u8XjctSNsA/edit#

Responding to Sage's possible kinks:

1. the cephbooter issue only comes up with new installs, existing vms with the files already downloaded are good -> ignoring for now ;)

2. ProxyPass lines are easy to change once we're happy with the new vms, the documented procedure allows you to leave the old vm running until that time.

3. NOPASSWD and ssh keys can be handled with downburst for new deployments; existing images are fine.

4. not installing debian images from scratch for now