Erlang Day Three

The final day of Erlang looks at Erlang’s core strength: concurrency. Erlang allows you to spawn processes that send and receive messages, both synchronously and asynchronously. Erlang’s processes can also be linked together — you can have a process monitoring another process and respawning it when it dies. This is a core idea in Erlang, and it’s why the language is so fault-tolerant.

The challenges for this day were interesting; but I only really got into the first one. The second challenge was to figure out how to make a process respawn itself when it dies, and after many hours researching and chatting to seasoned Erlang programmers, I’ve learned that this just isn’t a thing you do in Erlang.

  1. Things to find
  2. An OTP supervisor
  3. How to build an OTP server
  4. Things to do
  5. Restart a process
  6. Thoughts

Things to find

An OTP supervisor

Find an OTP service that will restart a process if it dies.

It looks as though you would use the supervisor behaviour with the one_for_one restart strategy from OTP for this. You can find some documentation and examples for this here.

The name OTP (Open Telecom Platform) is a bit misleading since it’s not about telecommunications. OTP is a bunch of libraries, conventions and tools to help write more robust Erlang applications, so even if you’re not writing Erlang for telecommunication apps you’ll probably still want to use OTP.

How to build an OTP server

In a very nice book called Learn You Some Erlang for Great Good! (free to read online), author Fred Hebert provides an example of a basic OTP server.

Things to do

Restart a process

Monitor the translate_service and restart it should it die.

Without getting into too much detail, we can see this code sets up an infinite loop and uses pattern matching to receive two different Spanish strings. If we send the service anything other than “casa” or “blanca”, the service tells us it doesn’t understand.

When a pattern is matched, the script sends the English string to the process ID that was unified with the From variable. Erlang uses the ! operator for message sending, so you can see where Scala borrowed some of its design ideas from.

It’s not immediately obvious, but the process is definitely killed when we send an unrecognised string. We can keep sending messages, but the translation service won’t respond to them. We can create a monitor function that starts another receive loop and watches the translation service for exit events.

The important line there is register(translator, spawn_link(fun loop/0)). The spawn_link call starts a new process, returns its PID, and links the new process to the one calling it. The register call creates an association in Erlang’s registry between the name translator and the PID that spawn_link returned.


There’s plenty to like about Erlang; I like the syntax, save for the different end-of-line characters. I like the way Erlang does concurrency; I’ve heard concurrency can be quite awkward but with Erlang it seemed painless. If at some point I work on a project that really needs concurrent processes, Erlang seems like the right tool to reach for.

I found the process of manually compiling modules cumbersome, so when I eventually return to Erlang I’d like to look into how to make a smoother workflow. I think the options for Erlang tooling is a different rabbit hole for a different day.