In OIL, we have a job scheduler that picks a set of machines (in MAAS), marks them internally as allocated, and then submits a job to jenkins, where juju runs, acquires the machines, and deploys stuff to them.
Between when the scheduler picks the machines and when juju starts deploying to them, which is often 20+ minutes, there is no indication to anyone that the machines are not available - they're still in the "Ready" state in MAAS.
We would like to scheduler to be able to acquire the machines via the MAAS API so other users can't acquire them (this is possible today), and for juju to be able to deploy to these already acquired machines (this is not).
This will make it much easier to share machines in the OIL lab, because users will be able to acquire any machine and use it without fear of picking a machine the scheduler has already internally allocated.
From discussion with Jason, our understanding is that the Juju MAAS provider uses a pattern of:
1. Allocate
2. Deploy
I've tried to figure out in the code if that's actually true but it's really hard to understand how it works. So:
- Is the above correct?
- If it is, could we move to just call Deploy and allocate just happens as a by-product?