Designing for Scalability with Erlang/OTP

Chapter 6. Finite State Machines

Now that we’ve become experts at writing generic servers, the time has come to master our next behavior. When prototyping systems with what eventually became Erlang, language inventors Joe Armstrong, Mike Williams, and Robert Virding were implementing a soft telephony switch allowing them to phone each other and say hello.¹ Each phone accessing the switch was prototyped as a process acting as an FSM. At any one time, the function would represent the state the phone was in (on hook, off hook, dialing, ringing, etc.) and receive events associated with that state (incoming call, dial, off hook, on hook, etc.).

One of the outcomes of this prototyping activity was to ensure that Erlang became a language suited for and optimized for building nontrivial and scalable FSMs, a key component in many complex systems. Developers use FSMs to program protocol stacks, connectors, proxies, workflow systems, gaming engines, and simulations, to mention but a few examples. So it was no surprise that when OTP behaviors came along, they included generic FSMs.

In this chapter, we introduce FSMs implemented in pure Erlang. We break an example up into generic and specific code, migrating it to the gen_fsm behavior. The good news is that all the borderline cases relating to concurrency and error handling that apply to generic servers also apply to FSMs. So while we might mention some of them, there will be no need for us to go into the same level of detail. After all, an ...