@ThreadSafe class StateMachineManager
A StateMachineManager is responsible for coordination and persistence of multiple FlowStateMachine objects. Each such object represents an instantiation of a (two-party) flow 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.
The SMM will always invoke the flow fibers on the given AffinityExecutor, regardless of which thread actually starts them via add.
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 can't hurt TODO: Implement stub/skel classes that provide a basic RPC framework on top of this.
Change |
data class Change |
FiberScheduler |
inner class FiberScheduler : FiberExecutorScheduler |
FlowSession |
data class FlowSession |
FlowSessionState |
sealed class FlowSessionState
FlowSessionState describes the session's state. |
<init> |
StateMachineManager(serviceHub: ServiceHubInternal, tokenizableServices: List<Any>, checkpointStorage: CheckpointStorage, executor: AffinityExecutor, database: Database, unfinishedFibers: ReusableLatch = ReusableLatch())
A StateMachineManager is responsible for coordination and persistence of multiple FlowStateMachine objects. Each such object represents an instantiation of a (two-party) flow that has reached a particular point. |
allStateMachines |
val allStateMachines: List<FlowLogic<*>> |
changes |
val changes: Observable<Change>
An observable that emits triples of the changing flow, the type of change, and a process-specific ID number which may change across restarts. |
checkpointStorage |
val checkpointStorage: CheckpointStorage |
database |
val database: Database |
executor |
val executor: AffinityExecutor |
scheduler |
val scheduler: FiberScheduler |
serviceHub |
val serviceHub: ServiceHubInternal |
add |
fun <T> add(logic: FlowLogic<T>): FlowStateMachine<T>
Kicks off a brand new state machine of the given class. The state machine will be persisted when it suspends, with automated restart if the StateMachineManager is restarted with checkpointed state machines in the storage service. |
findStateMachines |
fun <P : FlowLogic<T>, T> findStateMachines(flowClass: Class<P>): List<Pair<P, ListenableFuture<T>>>
Returns a list of all state machines executing the given flow logic at the top level (subflows do not count) |
start |
fun start(): Unit |
stop |
fun stop(allowedUnsuspendedFiberCount: Int = 0): Unit
Start the shutdown process, bringing the StateMachineManager to a controlled stop. When this method returns, all Fibers have been suspended and checkpointed, or have completed. |
track |
fun track(): Pair<List<FlowStateMachineImpl<*>>, Observable<Change>>
Atomic get snapshot + subscribe. This is needed so we don't miss updates between subscriptions to changes and calls to allStateMachines |