The Provisioning Agent (PA) will run inside a machine in an Ensemble environment, and will be responsible for starting new machines when necessary.
The agent must subclass the common base for agents in Ensemble, as defined in bug #614508.
A few requirements for this agent:
a) The PA must be architected in such a way that having multiple machine agents running against the same state is not a problem. For this, the PA must define lock points in which it won't try to do anything when another PA is already taking care of a updating the real deployment to match the requested state.
b) The PA must be able to be restarted at any point in time without side effects, even in the middle of an operation. As a corollary, the PA must, on start up, look for both entirely unhandled requests and also partial initialization states, and take any measures to enforce the desired final state in the real deployment (following (a) requirements).
c) The PA should monitor the running state and verify that a Machine Agent (MA, bug #617042) connects in an appropriate amount of time. If that doesn't happen, the PA should stop the virtual machine and attempt to fire a new one (again, considering (a) and (b) requirements). This behavior should be implemented under a separate branch, and is covered by bug #616565.
d) The PA should monitor the running state and verify that the MA continues to be connected. If the MA disconnects and doesn't reconnect in an expected amount of time, the machine should be shut down and restarted. This behavior is very similar to the one described in (c), and should be implemented under the same umbrella.
The work described on this ticket relies heavily on the machine state work defined in bug #616487, and items (c) and (d) which are scheduled for the future also depend on bug #616555 and bug #617046.
