Re: MSM and multithreading. Not clear what pattern to follow.

From: Richard Hodges (hodges.r_at_[hidden])
Date: 2021-03-08 15:11:09

On Mon, 8 Mar 2021 at 15:38, Andrea Fiorito via Boost-users <
boost-users_at_[hidden]> wrote:

> Hi, I'm trying to use MSM to implement a client server component for my
> app. The client is broadcasting a msg, that servers in a local network will
> detect and respond. The client at that point will connect to one of the
> servers and interact following the network protocol. For the network and
> threading I'm using boost ASIO.
>
> I have 2 submachines that manage one the discovering part, and the other
> the connection and dialogue with one server part. I found that would be
> beneficial to run the submachines in one thread managed by an io_context,
> with the idea of stopping that context to cancel all the asynchronous
> operations.
>
> Unfortunately there are some weird behaviours with submachines, and not
> getting if I missed the right way of doing things or else. the description
> of my last stopper is here:
>
> https://github.com/boostorg/msm/issues/37
>
> Is anyone able to point me on the right direction?
>

When I've done this in the past I've had most success by separating out the
thread/asio code from the state definitions and storing it in the class
that represents the state machine front end. All interaction with Asio
takes place by a call to a method on the FSM.

In this way you can insulate the state machine transitions from issues
caused by threads or asynchronous callbacks by ensuring that every call to
`process_event` happens in a function being executed by the strand.

It also eases state transitions if you don't make the state machine worry
about waiting for completion handlers to execute. You can keep a
"generation counter" or "token" in the handler so that it becomes a NO-OP
if the async operation is out of date.

A partial example on godbolt: https://godbolt.org/z/TzY17h

>
> Regards,
> Andrea
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> https://lists.boost.org/mailman/listinfo.cgi/boost-users
>

-- 
Richard Hodges
hodges.r_at_[hidden]
office: [phone]
home: [phone]
mobile: [phone]