Description of problem: When using automated scripts for deployment, many wait for sshd to come up, then ssh in. Since cloud-init and sshd are started in parallel, this creates a race condition for cloud-init to add ssh keys before sshd starts or the user can't login and the automated scripts can fail. Specifically, this is happening to me using test-kitchen with the kitchen-openstack plugin, which uses Fog. It calls wait_for and watches for sshd to come up. It catches sshd before cloud-init finishes installing keys, and fails to ssh. Version-Release number of selected component (if applicable): 0.7.2-7.fc20 Reproducing: Attempt to ssh in before cloud-init finishes but after sshd is up and running. Steps to Reproduce: 1. Pull in Fedora Cloud image for OpenStack 2. Configure test kitchen to use Fedora 3. Run test-kitchen tests Actual results: ssh fails, which causes test-kitchen or other automated scripts to fail. Expected results: ssh should succeed. Additional info:
I submitted a bug with upstream as well, because this is something that can affect other distributions.
This should be fixed in cloud-init-0.7.6-3.fc22. cloud-init-0.7.6-4.20140218bzr1060.fc22 will be pushed shortly with some major changes, so it may be worth testing against that instead.
Update: cloud-init-0.7.6-4.20140218bzr1060.fc22 isn't going to make it into F22 due to bug 1194577. If you're testing rawhide it will appear there, but otherwise feel free to go ahead and test cloud-init-0.7.6-3.fc22. Sorry about the confusion.