Tracing a build through OpenStack Compute (Nova)

My work at Rackspace has changed a bit in the last few weeks and I’ve shifted from managing a team of engineers to a full technical focus on OpenStack Nova. Although it was difficult to leave my management position, I’m happy to get back to my roots and dig into the technical stuff again.

One of the first things I wanted to tackle was understanding how a build request flows through Nova to a XenServer hypervisor. Following this process through the code is a bit tricky (I’m still learning python, so that could explain it). Here are the basic steps:

• Client requests a build via the API.
• The API runs some checks (quotas, auth, etc) and hands the build off to the scheduler.
• The scheduler figures out where the instance should go.
• The scheduler drops a message in queue specific to one compute node (where the instance will be built).
• The API responds to the client and the client is now unblocked and free to do other things.
• The compute node updates the database with the instance details and calls to the hypervisor to assemble block devices for the instance.
• A message is dropped into the network node’s queue (from the compute node) to begin assembling networks for the instance. The compute node blocks and waits while this step completes.
• Once the networking details come back (via the queue), the compute node does the remaining adjustments on the hypervisor and starts up the actual instance.
• When the instance starts successfully (or fails to do so), the database is updated and a message is dropped onto another message queue as a notification that the build is complete.

Click on the thumbnail on the right to see the flow chart I created to explain this process.

Please note: This information should be accurate to the Nova code as of November 1, 2011. There could be some refactoring of these build processes before Essex is released.