A higher level API for state management

[sajith]

Spinning this off #244.

There could be a TEManager.get_state() method that returns all the state we care about, and then SDX Controller would not have to care about details (topologies present, VLAN/bandwidth/latency allocations):

# At various SDX Controller checkpoints:
state = temanager.get_state()
db.upsert("temanager-state", state)

# During SDX Controller startup:
state = db.read("temanager-state")
temanager.restore_state(state)

Basically: when there are events that cause state to change (a topology is published, a new connection is made, a link was cut, service degraded... etc?), SDX Controller would have to call TEManager.get_state() and save the results in the DB. And then SDX Controller should call TEManager.restore_state() at startup, so that PCE can resume its work between SDX Controller restarts.

If PCE was a service, we probably could have done this in an even more transparent manner -- in that case PCE would get to maintain its state persistently (in a DB), but since PCE is only a library, SDX Controller will have to do the state management.

Thoughts, @YufengXin, @congwang09, @italovalcy?

[YufengXin]

We discussed making PCE a service earlier, too much hustle for the time.

But what you described can be considered a 'virtual" PCE service, just use sdx-controller + DB as the API front-:). I can see it should work fine.

It would be beneficial to formally define the TE manager state and state transition ontology, as we're trying to do for the connection state.

[YufengXin]

To ensure consistency and simplicity for sdx_controller, both available_bandwidth and vlan_range in port service are updated in the topology JSON objects that are saved in DB.

After crash recovery, the topology objects with the latest states will be recovered from the DB.