StateMachineManager(serviceHub: ServiceHub, runInThread: Executor)
A StateMachineManager is responsible for coordination and persistence of multiple ProtocolStateMachine objects. Each such object represents an instantiation of a (two-party) protocol that has reached a particular point.
An implementation of this class will persist state machines to long term storage so they can survive process restarts and, if run with a single-threaded executor, will ensure no two state machines run concurrently with each other (bad for performance, good for programmer mental health).
A "state machine" is a class with a single call method. The call method and any others it invokes are rewritten by a bytecode rewriting engine called Quasar, to ensure the code can be suspended and resumed at any point.
TODO: Session IDs should be set up and propagated automatically, on demand. TODO: Consider the issue of continuation identity more deeply: is it a safe assumption that a serialised continuation is always unique? TODO: Think about how to bring the system to a clean stop so it can be upgraded without any serialised stacks on disk TODO: Timeouts TODO: Surfacing of exceptions via an API and/or management UI TODO: Ability to control checkpointing explicitly, for cases where you know replaying a message cant hurt TODO: Make Kryo (de)serialize markers for heavy objects that are currently in the service hub. This avoids mistakes where services are temporarily put on the stack. TODO: Implement stub/skel classes that provide a basic RPC framework on top of this.