The current state of PCE (TEManager in particular) is not sufficient to offer concurrent and thread-safe handling of topologies and connections. Implementing #128 should address some of those concerns, but we should also make PCE asynchronous so that SDX Controller do not remain blocked waiting for PCE.
There are two possible options in making PCE asynchronous:
- Make PCE a stand-alone service with a REST API.
- Make PCE APIs async.
Either option possibly comes with a set of trade-offs. Option 1 can make things harder to deploy because we'll be adding yet another service, thus adding more latency than necessary and more failure-prone parts. Option 2 can make reasoning about things and implementation and debugging more complicated. It should be useful to do some designing and prototyping.
The current state of PCE (TEManager in particular) is not sufficient to offer concurrent and thread-safe handling of topologies and connections. Implementing #128 should address some of those concerns, but we should also make PCE asynchronous so that SDX Controller do not remain blocked waiting for PCE.
There are two possible options in making PCE asynchronous:
Either option possibly comes with a set of trade-offs. Option 1 can make things harder to deploy because we'll be adding yet another service, thus adding more latency than necessary and more failure-prone parts. Option 2 can make reasoning about things and implementation and debugging more complicated. It should be useful to do some designing and prototyping.