2016 Ericsson AB. All Rights Reserved. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. gen_statem Behavior statem.xml

This section is to be read with the gen_statem(3) manual page in STDLIB, where all interface functions and callback functions are described in detail.

This is a new behavior in Erlang/OTP 19.0. It has been thoroughly reviewed, is stable enough to be used by at least two heavy OTP applications, and is here to stay. Depending on user feedback, we do not expect but can find it necessary to make minor not backward compatible changes into Erlang/OTP 20.0.

Event-Driven State Machines

Established Automata Theory does not deal much with how a state transition is triggered, but assumes that the output is a function of the input (and the state) and that they are some kind of values.

For an Event-Driven State Machine, the input is an event that triggers a state transition and the output is actions executed during the state transition. It can analogously to the mathematical model of a Finite-State Machine be described as a set of relations of the following form:

State(S) x Event(E) -> Actions(A), State(S')

These relations are interpreted as follows: if we are in state S and event E occurs, we are to perform actions A and make a transition to state S'. Notice that S' can be equal to S.

As A and S' depend only on S and E, the kind of state machine described here is a Mealy Machine (see, for example, the corresponding Wikipedia article).

Like most gen_ behaviors, gen_statem keeps a server Data besides the state. Because of this, and as there is no restriction on the number of states (assuming that there is enough virtual machine memory) or on the number of distinct input events, a state machine implemented with this behavior is in fact Turing complete. But it feels mostly like an Event-Driven Mealy Machine.

Callback Modes

The gen_statem behavior supports two callback modes:

In mode state_functions, the state transition rules are written as some Erlang functions, which conform to the following convention:

StateName(EventType, EventContent, Data) ->
    .. code for actions here ...
    {next_state, NewStateName, NewData}.

In mode handle_event_function, only one Erlang function provides all state transition rules:

handle_event(EventType, EventContent, State, Data) ->
    .. code for actions here ...
    {next_state, NewState, NewData}

Both these modes allow other return tuples; see Module:StateName/3 in the gen_statem manual page. These other return tuples can, for example, stop the machine, execute state transition actions on the machine engine itself, and send replies.

Choosing the Callback Mode

The two callback modes give different possibilities and restrictions, but one goal remains: you want to handle all possible combinations of events and states.

This can be done, for example, by focusing on one state at the time and for every state ensure that all events are handled. Alternatively, you can focus on one event at the time and ensure that it is handled in every state. You can also use a mix of these strategies.

With state_functions, you are restricted to use atom-only states, and the gen_statem engine branches depending on state name for you. This encourages the callback module to gather the implementation of all event actions particular to one state in the same place in the code, hence to focus on one state at the time.

This mode fits well when you have a regular state diagram, like the ones in this chapter, which describes all events and actions belonging to a state visually around that state, and each state has its unique name.

With handle_event_function, you are free to mix strategies, as all events and states are handled in the same callback function.

This mode works equally well when you want to focus on one event at the time or on one state at the time, but function Module:handle_event/4 quickly grows too large to handle without branching to helper functions.

The mode enables the use of non-atom states, for example, complex states or even hierarchical states. If, for example, a state diagram is largely alike for the client side and the server side of a protocol, you can have a state {StateName,server} or {StateName,client}, and make StateName determine where in the code to handle most events in the state. The second element of the tuple is then used to select whether to handle special client-side or server-side events.

Example

This example starts off as equivalent to the example in section gen_fsm Behavior. In later sections, additions and tweaks are made using features in gen_statem that gen_fsm does not have. The end of this chapter provides the example again with all the added features.

A door with a code lock can be seen as a state machine. Initially, the door is locked. When someone presses a button, an event is generated. Depending on what buttons have been pressed before, the sequence so far can be correct, incomplete, or wrong. If correct, the door is unlocked for 10 seconds (10,000 milliseconds). If incomplete, we wait for another button to be pressed. If wrong, we start all over, waiting for a new button sequence.

Code Lock State Diagram

This code lock state machine can be implemented using gen_statem with the following callback module:

gen_statem:start_link({local,?NAME}, ?MODULE, Code, []). button(Digit) -> gen_statem:cast(?NAME, {button,Digit}). init(Code) -> do_lock(), Data = #{code => Code, remaining => Code}, {ok,locked,Data}. callback_mode() -> state_functions. locked( cast, {button,Digit}, #{code := Code, remaining := Remaining} = Data) -> case Remaining of [Digit] -> do_unlock(), {next_state,open,Data#{remaining := Code},10000}; [Digit|Rest] -> % Incomplete {next_state,locked,Data#{remaining := Rest}}; _Wrong -> {next_state,locked,Data#{remaining := Code}} end. open(timeout, _, Data) -> do_lock(), {next_state,locked,Data}; open(cast, {button,_}, Data) -> do_lock(), {next_state,locked,Data}. do_lock() -> io:format("Lock~n", []). do_unlock() -> io:format("Unlock~n", []). terminate(_Reason, State, _Data) -> State =/= locked andalso do_lock(), ok. code_change(_Vsn, State, Data, _Extra) -> {ok,State,Data}. ]]>

The code is explained in the next sections.

Starting gen_statem

In the example in the previous section, gen_statem is started by calling code_lock:start_link(Code):

gen_statem:start_link({local,?NAME}, ?MODULE, Code, []). ]]>

start_link calls function gen_statem:start_link/4, which spawns and links to a new process, a gen_statem.

The first argument, {local,?NAME}, specifies the name. In this case, the gen_statem is locally registered as code_lock through the macro ?NAME.

If the name is omitted, the gen_statem is not registered. Instead its pid must be used. The name can also be specified as {global,Name}, then the gen_statem is registered using global:register_name/2 in Kernel.

The second argument, ?MODULE, is the name of the callback module, that is, the module where the callback functions are located, which is this module.

The interface functions (start_link/1 and button/1) are located in the same module as the callback functions (init/1, locked/3, and open/3). It is normally good programming practice to have the client-side code and the server-side code contained in one module.

The third argument, Code, is a list of digits, which is the correct unlock code that is passed to callback function init/1.

The fourth argument, [], is a list of options. For the available options, see gen_statem:start_link/3.

If name registration succeeds, the new gen_statem process calls callback function code_lock:init(Code). This function is expected to return {ok,State,Data}, where State is the initial state of the gen_statem, in this case locked; assuming that the door is locked to begin with. Data is the internal server data of the gen_statem. Here the server data is a map with key code that stores the correct button sequence, and key remaining that stores the remaining correct button sequence (the same as the code to begin with).

do_lock(), Data = #{code => Code, remaining => Code}, {ok,locked,Data}. ]]>

Function gen_statem:start_link is synchronous. It does not return until the gen_statem is initialized and is ready to receive events.

Function gen_statem:start_link must be used if the gen_statem is part of a supervision tree, that is, started by a supervisor. Another function, gen_statem:start can be used to start a standalone gen_statem, that is, a gen_statem that is not part of a supervision tree.

state_functions. ]]>

Function Module:callback_mode/0 selects the CallbackMode for the callback module, in this case state_functions. That is, each state has got its own handler function.

Handling Events

The function notifying the code lock about a button event is implemented using gen_statem:cast/2:

gen_statem:cast(?NAME, {button,Digit}). ]]>

The first argument is the name of the gen_statem and must agree with the name used to start it. So, we use the same macro ?NAME as when starting. {button,Digit} is the event content.

The event is made into a message and sent to the gen_statem. When the event is received, the gen_statem calls StateName(cast, Event, Data), which is expected to return a tuple {next_state,NewStateName,NewData}. StateName is the name of the current state and NewStateName is the name of the next state to go to. NewData is a new value for the server data of the gen_statem.

case Remaining of [Digit] -> % Complete do_unlock(), {next_state,open,Data#{remaining := Code},10000}; [Digit|Rest] -> % Incomplete {next_state,locked,Data#{remaining := Rest}}; [_|_] -> % Wrong {next_state,locked,Data#{remaining := Code}} end. open(timeout, _, Data) -> do_lock(), {next_state,locked,Data}; open(cast, {button,_}, Data) -> do_lock(), {next_state,locked,Data}. ]]>

If the door is locked and a button is pressed, the pressed button is compared with the next correct button. Depending on the result, the door is either unlocked and the gen_statem goes to state open, or the door remains in state locked.

If the pressed button is incorrect, the server data restarts from the start of the code sequence.

In state open, any button locks the door, as any event cancels the event timer, so no time-out event occurs after a button event.

Event Time-Outs

When a correct code has been given, the door is unlocked and the following tuple is returned from locked/2:

10,000 is a time-out value in milliseconds. After this time (10 seconds), a time-out occurs. Then, StateName(timeout, 10000, Data) is called. The time-out occurs when the door has been in state open for 10 seconds. After that the door is locked again:

do_lock(), {next_state,locked,Data}; ]]>
All State Events

Sometimes events can arrive in any state of the gen_statem. It is convenient to handle these in a common state handler function that all state functions call for events not specific to the state.

Consider a code_length/0 function that returns the length of the correct code (that should not be sensitive to reveal). We dispatch all events that are not state-specific to the common function handle_event/3:

gen_statem:call(?NAME, code_length). ... locked(...) -> ... ; locked(EventType, EventContent, Data) -> handle_event(EventType, EventContent, Data). ... open(...) -> ... ; open(EventType, EventContent, Data) -> handle_event(EventType, EventContent, Data). handle_event({call,From}, code_length, #{code := Code} = Data) -> {keep_state,Data,[{reply,From,length(Code)}]}. ]]>

This example uses gen_statem:call/2, which waits for a reply from the server. The reply is sent with a {reply,From,Reply} tuple in an action list in the {keep_state,...} tuple that retains the current state.

One Event Handler

If mode handle_event_function is used, all events are handled in Module:handle_event/4 and we can (but do not have to) use an event-centered approach where we first branch depending on event and then depending on state:

handle_event_function. handle_event(cast, {button,Digit}, State, #{code := Code} = Data) -> case State of locked -> case maps:get(remaining, Data) of [Digit] -> % Complete do_unlock(), {next_state,open,Data#{remaining := Code},10000}; [Digit|Rest] -> % Incomplete {keep_state,Data#{remaining := Rest}}; [_|_] -> % Wrong {keep_state,Data#{remaining := Code}} end; open -> do_lock(), {next_state,locked,Data} end; handle_event(timeout, _, open, Data) -> do_lock(), {next_state,locked,Data}. ... ]]>
Stopping
In a Supervision Tree

If the gen_statem is part of a supervision tree, no stop function is needed. The gen_statem is automatically terminated by its supervisor. Exactly how this is done is defined by a shutdown strategy set in the supervisor.

If it is necessary to clean up before termination, the shutdown strategy must be a time-out value and the gen_statem must in function init/1 set itself to trap exit signals by calling process_flag(trap_exit, true):

process_flag(trap_exit, true), do_lock(), ... ]]>

When ordered to shut down, the gen_statem then calls callback function terminate(shutdown, State, Data).

In this example, function terminate/3 locks the door if it is open, so we do not accidentally leave the door open when the supervision tree terminates:

State =/= locked andalso do_lock(), ok. ]]>
Standalone gen_statem

If the gen_statem is not part of a supervision tree, it can be stopped using gen_statem:stop, preferably through an API function:

gen_statem:stop(?NAME). ]]>

This makes the gen_statem call callback function terminate/3 just like for a supervised server and waits for the process to terminate.

Actions

In the first sections actions were mentioned as a part of the general state machine model. These general actions are implemented with the code that callback module gen_statem executes in an event-handling callback function before returning to the gen_statem engine.

There are more specific state-transition actions that a callback function can order the gen_statem engine to do after the callback function return. These are ordered by returning a list of actions in the return tuple from the callback function. These state transition actions affect the gen_statem engine itself and can do the following:

Postpone the current event Hibernate the gen_statem Start an event time-out Reply to a caller Generate the next event to handle

In the example earlier was mentioned the event time-out and replying to a caller. An example of event postponing is included later in this chapter. For details, see the gen_statem(3) manual page. You can, for example, reply to many callers and generate multiple next events to handle.

Event Types

The previous sections mentioned a few event types. Events of all types are handled in the same callback function, for a given state, and the function gets EventType and EventContent as arguments.

The following is a complete list of event types and where they come from:

cast Generated by gen_statem:cast. {call,From} Generated by gen_statem:call, where From is the reply address to use when replying either through the state transition action {reply,From,Msg} or by calling gen_statem:reply. info Generated by any regular process message sent to the gen_statem process. timeout Generated by state transition action {timeout,Time,EventContent} (or its short form Time) timer timing out. internal Generated by state transition action {next_event,internal,EventContent}. All event types above can also be generated using {next_event,EventType,EventContent}.
State Time-Outs

The time-out event generated by state transition action {timeout,Time,EventContent} is an event time-out, that is, if an event arrives the timer is cancelled. You get either an event or a time-out, but not both.

Often you want a timer not to be cancelled by any event or you want to start a timer in one state and respond to the time-out in another. This can be accomplished with a regular Erlang timer: erlang:start_timer.

For the example so far in this chapter: using the gen_statem event timer has the consequence that if a button event is generated while in the open state, the time-out is cancelled and the button event is delivered. So, we choose to lock the door if this occurred.

Suppose that we do not want a button to lock the door, instead we want to ignore button events in the open state. Then we start a timer when entering the open state and wait for it to expire while ignoring button events:

case Remaining of [Digit] -> do_unlock(), Tref = erlang:start_timer(10000, self(), lock), {next_state,open,Data#{remaining := Code, timer := Tref}}; ... open(info, {timeout,Tref,lock}, #{timer := Tref} = Data) -> do_lock(), {next_state,locked,Data}; open(cast, {button,_}, Data) -> {keep_state,Data}; ... ]]>

If you need to cancel a timer because of some other event, you can use erlang:cancel_timer(Tref). Notice that a time-out message cannot arrive after this, unless you have postponed it (see the next section) before, so ensure that you do not accidentally postpone such messages.

Another way to cancel a timer is not to cancel it, but to ignore it if it arrives in a state where it is known to be late.

Postponing Events

If you want to ignore a particular event in the current state and handle it in a future state, you can postpone the event. A postponed event is retried after the state has changed, that is, OldState =/= NewState.

Postponing is ordered by the state transition action postpone.

In this example, instead of ignoring button events while in the open state, we can postpone them and they are queued and later handled in the locked state:

{keep_state,Data,[postpone]}; ... ]]>

The fact that a postponed event is only retried after a state change translates into a requirement on the event and state space. If you have a choice between storing a state data item in the State or in the Data: if a change in the item value affects which events that are handled, then this item is to be part of the state.

You want to avoid that you maybe much later decide to postpone an event in one state and by misfortune it is never retried, as the code only changes the Data but not the State.

Fuzzy State Diagrams

It is not uncommon that a state diagram does not specify how to handle events that are not illustrated in a particular state in the diagram. Hopefully this is described in an associated text or from the context.

Possible actions: ignore as in drop the event (maybe log it) or deal with the event in some other state as in postpone it.

Selective Receive

Erlang's selective receive statement is often used to describe simple state machine examples in straightforward Erlang code. The following is a possible implementation of the first example:

spawn( fun () -> true = register(?NAME, self()), do_lock(), locked(Code, Code) end). button(Digit) -> ?NAME ! {button,Digit}. locked(Code, [Digit|Remaining]) -> receive {button,Digit} when Remaining =:= [] -> do_unlock(), open(Code); {button,Digit} -> locked(Code, Remaining); {button,_} -> locked(Code, Code) end. open(Code) -> receive after 10000 -> do_lock(), locked(Code, Code) end. do_lock() -> io:format("Locked~n", []). do_unlock() -> io:format("Open~n", []). ]]>

The selective receive in this case causes implicitly open to postpone any events to the locked state.

A selective receive cannot be used from a gen_statem behavior as for any gen_* behavior, as the receive statement is within the gen_* engine itself. It must be there because all sys compatible behaviors must respond to system messages and therefore do that in their engine receive loop, passing non-system messages to the callback module.

The state transition action postpone is designed to model selective receives. A selective receive implicitly postpones any not received events, but the postpone state transition action explicitly postpones one received event.

Both mechanisms have the same theoretical time and memory complexity, while the selective receive language construct has smaller constant factors.

Self-Generated Events

It can sometimes be beneficial to be able to generate events to your own state machine. This can be done with the state transition action {next_event,EventType,EventContent}.

You can generate events of any existing type, but the internal type can only be generated through action next_event. Hence, it cannot come from an external source, so you can be certain that an internal event is an event from your state machine to itself.

One example of using self-generated events can be when you have a state machine specification that uses state entry actions. You can code that using a dedicated function to do the state transition. But if you want that code to be visible besides the other state logic, you can insert an internal event that does the entry actions. This has the same unfortunate consequence as using state transition functions: everywhere you go to the state, you must explicitly insert the internal event or use a state transition function.

The following is an implementation of entry actions using internal events with content enter using a helper function enter/3 for state entry:

process_flag(trap_exit, true), Data = #{code => Code}, enter(ok, locked, Data). callback_mode() -> state_functions. locked(internal, enter, _Data) -> do_lock(), {keep_state,Data#{remaining => Code}}; locked( cast, {button,Digit}, #{code := Code, remaining := Remaining} = Data) -> case Remaining of [Digit] -> enter(next_state, open, Data); ... open(internal, enter, _Data) -> Tref = erlang:start_timer(10000, self(), lock), do_unlock(), {keep_state,Data#{timer => Tref}}; open(info, {timeout,Tref,lock}, #{timer := Tref} = Data) -> enter(next_state, locked, Data); ... enter(Tag, State, Data) -> {Tag,State,Data,[{next_event,internal,enter}]}. ]]>
Example Revisited

This section includes the example after all mentioned modifications and some more using the entry actions, which deserves a new state diagram:

Code Lock State Diagram Revisited

Notice that this state diagram does not specify how to handle a button event in the state open. So, you need to read somewhere else that unspecified events must be ignored as in not consumed but handled in some other state. Also, the state diagram does not show that the code_length/0 call must be handled in every state.

Callback Mode: state_functions

Using state functions:

gen_statem:start_link({local,?NAME}, ?MODULE, Code, []). stop() -> gen_statem:stop(?NAME). button(Digit) -> gen_statem:cast(?NAME, {button,Digit}). code_length() -> gen_statem:call(?NAME, code_length). init(Code) -> process_flag(trap_exit, true), Data = #{code => Code}, enter(ok, locked, Data). callback_mode() -> state_functions. locked(internal, enter, #{code := Code} = Data) -> do_lock(), {keep_state,Data#{remaining => Code}}; locked( cast, {button,Digit}, #{code := Code, remaining := Remaining} = Data) -> case Remaining of [Digit] -> % Complete enter(next_state, open, Data); [Digit|Rest] -> % Incomplete {keep_state,Data#{remaining := Rest}}; [_|_] -> % Wrong {keep_state,Data#{remaining := Code}} end; locked(EventType, EventContent, Data) -> handle_event(EventType, EventContent, Data). open(internal, enter, Data) -> Tref = erlang:start_timer(10000, self(), lock), do_unlock(), {keep_state,Data#{timer => Tref}}; open(info, {timeout,Tref,lock}, #{timer := Tref} = Data) -> enter(next_state, locked, Data); open(cast, {button,_}, _) -> {keep_state_and_data,[postpone]}; open(EventType, EventContent, Data) -> handle_event(EventType, EventContent, Data). handle_event({call,From}, code_length, #{code := Code}) -> {keep_state_and_data,[{reply,From,length(Code)}]}. enter(Tag, State, Data) -> {Tag,State,Data,[{next_event,internal,enter}]}. do_lock() -> io:format("Locked~n", []). do_unlock() -> io:format("Open~n", []). terminate(_Reason, State, _Data) -> State =/= locked andalso do_lock(), ok. code_change(_Vsn, State, Data, _Extra) -> {ok,State,Data}. ]]>
Callback Mode: handle_event_function

This section describes what to change in the example to use one handle_event/4 function. The previously used approach to first branch depending on event does not work that well here because of the generated entry actions, so this example first branches depending on state:

handle_event_function. %% State: locked handle_event(internal, enter, locked, #{code := Code} = Data) -> do_lock(), {keep_state,Data#{remaining => Code}}; handle_event( cast, {button,Digit}, locked, #{code := Code, remaining := Remaining} = Data) -> case Remaining of [Digit] -> % Complete enter(next_state, open, Data); [Digit|Rest] -> % Incomplete {keep_state,Data#{remaining := Rest}}; [_|_] -> % Wrong {keep_state,Data#{remaining := Code}} end; %% %% State: open handle_event(internal, enter, open, Data) -> Tref = erlang:start_timer(10000, self(), lock), do_unlock(), {keep_state,Data#{timer => Tref}}; handle_event(info, {timeout,Tref,lock}, open, #{timer := Tref} = Data) -> enter(next_state, locked, Data); handle_event(cast, {button,_}, open, _) -> {keep_state_and_data,[postpone]}; %% %% Any state handle_event({call,From}, code_length, _State, #{code := Code}) -> {keep_state_and_data,[{reply,From,length(Code)}]}. ... ]]>

Notice that postponing buttons from the locked state to the open state feels like the wrong thing to do for a code lock, but it at least illustrates event postponing.

Filter the State

The example servers so far in this chapter print the full internal state in the error log, for example, when killed by an exit signal or because of an internal error. This state contains both the code lock code and which digits that remain to unlock.

This state data can be regarded as sensitive, and maybe not what you want in the error log because of some unpredictable event.

Another reason to filter the state can be that the state is too large to print, as it fills the error log with uninteresting details.

To avoid this, you can format the internal state that gets in the error log and gets returned from sys:get_status/1,2 by implementing function Module:format_status/2, for example like this:

StateData = {State, maps:filter( fun (code, _) -> false; (remaining, _) -> false; (_, _) -> true end, Data)}, case Opt of terminate -> StateData; normal -> [{data,[{"State",StateData}]}] end. ]]>

It is not mandatory to implement a Module:format_status/2 function. If you do not, a default implementation is used that does the same as this example function without filtering the Data term, that is, StateData = {State,Data}, in this example containing sensitive information.

Complex State

The callback mode handle_event_function enables using a non-atom state as described in section Callback Modes, for example, a complex state term like a tuple.

One reason to use this is when you have a state item that affects the event handling, in particular in combination with postponing events. We complicate the previous example by introducing a configurable lock button (this is the state item in question), which in the open state immediately locks the door, and an API function set_lock_button/1 to set the lock button.

Suppose now that we call set_lock_button while the door is open, and have already postponed a button event that until now was not the lock button. The sensible thing can be to say that the button was pressed too early so it is not to be recognized as the lock button. However, then it can be surprising that a button event that now is the lock button event arrives (as retried postponed) immediately after the state transits to locked.

So we make the button/1 function synchronous by using gen_statem:call and still postpone its events in the open state. Then a call to button/1 during the open state does not return until the state transits to locked, as it is there the event is handled and the reply is sent.

If a process now calls set_lock_button/1 to change the lock button while another process hangs in button/1 with the new lock button, it can be expected that the hanging lock button call immediately takes effect and locks the lock. Therefore, we make the current lock button a part of the state, so that when we change the lock button, the state changes and all postponed events are retried.

We define the state as {StateName,LockButton}, where StateName is as before and LockButton is the current lock button:

gen_statem:start_link( {local,?NAME}, ?MODULE, {Code,LockButton}, []). stop() -> gen_statem:stop(?NAME). button(Digit) -> gen_statem:call(?NAME, {button,Digit}). code_length() -> gen_statem:call(?NAME, code_length). set_lock_button(LockButton) -> gen_statem:call(?NAME, {set_lock_button,LockButton}). init({Code,LockButton}) -> process_flag(trap_exit, true), Data = #{code => Code, remaining => undefined, timer => undefined}, enter(ok, {locked,LockButton}, Data, []). callback_mode() -> handle_event_function. handle_event( {call,From}, {set_lock_button,NewLockButton}, {StateName,OldLockButton}, Data) -> {next_state,{StateName,NewLockButton},Data, [{reply,From,OldLockButton}]}; handle_event( {call,From}, code_length, {_StateName,_LockButton}, #{code := Code}) -> {keep_state_and_data, [{reply,From,length(Code)}]}; handle_event( EventType, EventContent, {locked,LockButton}, #{code := Code, remaining := Remaining} = Data) -> case {EventType,EventContent} of {internal,enter} -> do_lock(), {keep_state,Data#{remaining := Code}}; {{call,From},{button,Digit}} -> case Remaining of [Digit] -> % Complete next_state( {open,LockButton}, Data, [{reply,From,ok}]); [Digit|Rest] -> % Incomplete {keep_state,Data#{remaining := Rest}, [{reply,From,ok}]}; [_|_] -> % Wrong {keep_state,Data#{remaining := Code}, [{reply,From,ok}]} end end; handle_event( EventType, EventContent, {open,LockButton}, #{timer := Timer} = Data) -> case {EventType,EventContent} of {internal,enter} -> Tref = erlang:start_timer(10000, self(), lock), do_unlock(), {keep_state,Data#{timer := Tref}}; {info,{timeout,Timer,lock}} -> next_state({locked,LockButton}, Data, []); {{call,From},{button,Digit}} -> if Digit =:= LockButton -> erlang:cancel_timer(Timer), next_state( {locked,LockButton}, Data, [{reply,From,locked}]); true -> {keep_state_and_data, [postpone]} end end. next_state(State, Data, Actions) -> enter(next_state, State, Data, Actions). enter(Tag, State, Data, Actions) -> {Tag,State,Data,[{next_event,internal,enter}|Actions]}. do_lock() -> io:format("Locked~n", []). do_unlock() -> io:format("Open~n", []). terminate(_Reason, State, _Data) -> State =/= locked andalso do_lock(), ok. code_change(_Vsn, State, Data, _Extra) -> {ok,State,Data}. format_status(Opt, [_PDict,State,Data]) -> StateData = {State, maps:filter( fun (code, _) -> false; (remaining, _) -> false; (_, _) -> true end, Data)}, case Opt of terminate -> StateData; normal -> [{data,[{"State",StateData}]}] end. ]]>

It can be an ill-fitting model for a physical code lock that the button/1 call can hang until the lock is locked. But for an API in general it is not that strange.

Hibernation

If you have many servers in one node and they have some state(s) in their lifetime in which the servers can be expected to idle for a while, and the amount of heap memory all these servers need is a problem, then the memory footprint of a server can be mimimized by hibernating it through proc_lib:hibernate/3.

It is rather costly to hibernate a process; see erlang:hibernate/3. It is not something you want to do after every event.

We can in this example hibernate in the {open,_} state, because what normally occurs in that state is that the state time-out after a while triggers a transition to {locked,_}:

case {EventType,EventContent} of {internal,enter} -> Tref = erlang:start_timer(10000, self(), lock), do_unlock(), {keep_state,Data#{timer := Tref},[hibernate]}; ... ]]>

The [hibernate] action list on the last line when entering the {open,_} state is the only change. If any event arrives in the {open,_}, state, we do not bother to rehibernate, so the server stays awake after any event.

To change that we would need to insert action hibernate in more places. For example, for the state-independent set_lock_button and code_length operations that then would have to be aware of using hibernate while in the {open,_} state, which would clutter the code.

This server probably does not use heap memory worth hibernating for. To gain anything from hibernation, your server would have to produce some garbage during callback execution, for which this example server can serve as a bad example.