Re: Sending Messages to Workers

Welcome Joseph,

It sounds like you're trying to roll your own process registry.

For simplicity's sake it may be a good idea to use something along the
lines of gproc or syn.

rough untested code

-define(PROC_NAME(X), {n, l, X}).
-define(VIA(X), {via, gproc, ?PROC_NAME(X)}).

start_link(Id) ->
  gen_server:start_link(?VIA(Id), ?MODULE, [Id], []).

init([Id]) ->
 ....
{ok, State}

and you can then call via gproc and let it do the work for you

gen_server:call(?VIA(Id), {msg, Val})

Kind regards,
Leonard

On Mon, Nov 15, 2021 at 2:08 PM Joseph Lloyd <[email protected]> wrote:
>
> I am new to Erlang and using mailing lists, so please forgive my poor etiquette, and feel free to correct me.
>
> Background:
> I am prototyping a simple application using Erlang.  The application has a manager that reads messages from a socket (UDP in this case) and then sends the message (based on a destination byte) to a handler process.  There is a possibility of 2^n handler processes. The number is determined dynamically at run time based on user input.
>
> A handler process is created when the manager process receives a message to create the handler. create/3 is part of the handler module's API and is essentially a wrapper around handler_sup:start_child/2.  When the manager process receives a packet, it sends a message containing the packet to the appropriate handler process.  The manager process uses an ets table for mapping destinations to handler pids.
>
> I implemented the handlers as gen_servers and created a handler_sup that will restart the handler processes whenever they go down.
>
> Problem:
> If a handler process goes down, the destination/pid mapping in the manager module becomes invalid.  The handler supervisor will generate a new process, but the manager doesn't know the pid for the replacement process.
>
> Question:
> What is the proper way to design this in Erlang?  Do I put the destination/pid mapping in the handler_sup module and query that whenever the manager wants to send the message? Or does that violate supervisors only being supervisors paradigm?
>
> Thanks in advance,
>
> Joseph