A well defined DSL for constructing finite state machines for use in Filecoin.
Wikipedia - Finite State Machines Wikipedia - UML State Machine
This library provides a way to model a Filecoin process that operates on a specific data structure as a series of state transitions. It is used in the storage and retrieval markets in go-fil-markets repository to model the lifecycle of a storage or retrieval deal. It may be used for other parts of Filecoin in the future.
A state machine is defined in terms of
StateType-- the type of data structure we are tracking (should be a struct)StateKeyField-- the field in the data structure that represents the unique identifier for the current state. Must be a field that is comparable in golang (i.e. not an array or map)Events-- Events are the list of inputs that trigger transitions in state. An event is defined in terms of:- An identifier
- A mapping between source states and destination states -- events can apply to one, many, or all source states. When there is more than one source state, the source states can share or have different destination states
- An
Action- this a function that applies updates to the underlying data structure in fields that are notStateKeyField
StateEntryFuncs- State entry functions are handlers that get called when the FSM enters a specific state. Where Actions are associated with specific events, state entry funcs are associated with specific states. Actions can modify the underlying data structure. StateEntryFuncs can only trigger additional events if they wish to modify the state (a state entry func receives a dereferenced value for the data structure, rather than the pointer)Environment- This is a single interface that is used to access external dependencies. It is available to aStateEntryFuncNotifier- A function that gets called on each successfully applied state transition, before any state entry func is called. This is useful for providing external notifications about state updates. It is called with the name of the event applied and the current state of the data structure.FinalityStates- a list of states from which the state machine cannot leave. When the statemachine enters these states, it shuts down and stops receiving messages.
Let's consider a hypothetical deal we want to track in Filecoin. Each deal will have a series of states it can be in, and various things that can happen to change its state. We will track multiple deals at the same time. In this example, we will model the actions of the receiving party (i.e. the person who accepted the deal and is responding)
Here' is the simplified deal structure:
type DealState struct {
// the original proposal
DealProposal
// the current status identifier
Status DealStatus
// the other party's network address
Receiver peer.ID
// how much much we received
TotalSent uint64
// how much money we have
FundsReceived abi.TokenAmount
// an informational message to augment the status of the deal
Message string
}Let's pretend our ideal deal flow looks like:
Receive new deal proposal -> Validate proposal -> For All Data Requested, Send a chunk, then request payment, then wait for payment before sending more -> Complete deal
You can imagine at each state in this happy path, things could go wrong. A deal proposal might need to be rejected for not meeting criteria. A network message could fail to send. A client could send only a partial payment, or one that fails to process. We might had an error reading our own data.
You can start to assemble different events that might happen in this process:
ReceivedNewDeal
AcceptedDeal
RejectedDeal
SentData
RequestedPayment
ReceivedPayment
...
And you can imagine different states the deal is in:
New
Accepted
Rejected
AwaitingPayment
Completed
Failed
...
We would track these states in the Status field of our deal.
So we have StateType - DealState - and a StateKeyField - Status. Now we need to define our events. Here's how we do that, using the modules custom DSL:
var DealEvents = fsm.Events{
fsm.Event("ReceivedNewDeal").FromAny().To("New").Action(func (deal * DealState, proposal DealProposal) error {
deal.DealProposal = proposal
return nil
}),
fsm.Event("RejectedDeal").From("New").To("Rejected").Action(func (deal * DealState, reason string) error {
deal.Message = fmt.Sprintf("Rejected deal because: %s", reason)
return nil
}),
fsm.Event("AcceptedDeal").From("New").To("Accepted"),
...
}As we enter each new state, there are things we need to do to advance the deal flow. When we have a new deal proposal, we need to validate it, to accept or reject. When we accept a deal, we need to start sending data. This is handled by StateEntryFuncs. We define this as a mapping between states and their handlers:
// assuming ValidateDeal, SendData, and SendRejection are defined elsewhere
var DealEntryFuncs = fsm.StateEntryFuncs{
"New": ValidateDeal,
"Rejected": SendRejection
"Accepted": SendData,
}As a baseline rule, where Actions on Events modify state, StateEntryFuncs leave state as it is but can have side effects. We therefore define an Environment type used to trigger these side effects with the outside world. We might define our environment here as:
type DealResponse struct {
Data: []byte
}
type DealEnvironment interface {
LookUpSomeData(proposal DealProposal) ([]byte, err)
SendDealResponse(response DealResponse) error
...
}And then our SendData StateEntryFunc function might look like this:
func SendData(ctx fsm.Context, environment DealEnvironment, deal DealState) error {
data, err := environment.LookupSomeData(deal.DealProposal)
if err != nil {
return ctx.Trigger("DataError", err)
}
err := environment.SendDealResponse(DealResponse{Data: data})
if err != nil {
return ctx.Trigger("NetworkError", err)
}
return ctx.Trigger("SentData", len(data))
}You can see our SendData interacts with the environment dispatches additional events using the Trigger function on the FSM Context which is also supplied to a StateEntryFunc. While the above function only dispatches at most one new event based on the conditional logic, it is possible to dismatch multiple events in a state entry func and they will be processed asynchronously in the order they are received.
Putting all this together, our state machine parameters are as follows:
var DealFSMParameters = fsm.Parameters{
StateType: DealState{},
StateKeyField: "Status",
Events: DealEvents,
StateEntryFuncs: DealEntryFuncs,
Environment: DealEnvironmentImplementation{}
FinalityStates: []StateKey{"Completed", "Failed"}
}The FSM module is designed to be used with a list of data structures, persisted to a datastore through the go-statestore module.
You initialize a new set of FSM's as follows:
var ds datastore.Batching
var dealStateMachines = fsm.New(ds, DealFSMParameters)You can now dispatch events from the outside to a particular deal FSM with:
var DealID // some identifier of a deal
var proposal DealProposal
dealStateMachines.Send(DealID, "ReceivedNewDeal", proposal)The statemachine will initialize a new FSM if it is not already tracking data for the given identifier (the first parameter -- in this case a deal ID)
That's it! The FSM module:
- persists state machines to disk with
go-statestore - operated asynchronously and in nonblocking fashion (any event dispatches will return immediate)
- operates multiple FSMs in parallel (each machine has its own go-routine)
This module is dual-licensed under Apache 2.0 and MIT terms.
Copyright 2019. Protocol Labs, Inc.