Registry and multi_call / abcast

I have a Registry to create dynamic GenServers (each holding users’ parameters). I want to be able to take advantage of GenServer’s distributed multi_call / abcast, but since I’m using Registry, I must use a via tuple. Seems that mutli_call doesn’t like via tuples… Am I doing something wrong, is this by design, bug, or…? Thanks!

  def start_link(uid, pid, params \\ %{}) do
    GenServer.start_link(__MODULE__, [{pid, params}], name: via_tuple(uid))

  defp via_tuple(uid), do: {:via, Registry, {@users_registry_name, uid}}
  def get_params(uid) do
    {replies, _bad} = via_tuple(uid) |> GenServer.multi_call(:get_params)

Error returned:

** (FunctionClauseError) no function clause matching in :gen_server.start_monitor/2
        (stdlib) gen_server.erl:576: :gen_server.start_monitor(:node1@host, {:via, Registry, {:users_registry, "test"}})
        (stdlib) gen_server.erl:467: :gen_server.send_nodes/5
        (stdlib) gen_server.erl:425: :gen_server.do_multi_call/4
(project_name) lib/project_name/users_monitor.ex:17: ProjectName.UsersMonitor.get_params/1

Digging through Erlang code for gen_server I see that start_monitor only accepts an atom as the name…

Any ideas how I can do something similar with as much ease as multi_call / abcast, or will I have to implement something myself?

multi_call and abcast are about sending calls/casts across nodes rather than on the same machine.

Right, so my question is - how do I create dynamically-named GenServers and cast to them across multiple nodes?
What I mean is - I could have multiple GenServers with the same dynamically-generated name on different nodes. I want to be able to retrieve all params I’ve saved on those GenServers regardless of what node they are on.

I think you’re looking for a process registry that works in a distributed manner. Registry is only a local process registry, so maybe it’s not the right choice for you?

Maybe swarm or syn are a better match for your case.

1 Like

Thanks! I was thinking about Swarm actually, but was wondering if there was anything native that I could use instead. Also - do names in Swarm have to be unique? If I start two GenServers with the same name, will sending to that name send to both?

Swarm supports process grouping, that might do the trick?

Hmm… I’m not sure that it would. My scenario is this: a user connects with certain params, and I need to save those params. The same user can connect multiple times, with either similar or different params, but what is constant is his user_id (which in this case would be the “group”). I need to be able to retrieve all params saved under that user_id. My problem is that I could have many users connecting, so I can’t use a dynamically-created atom from the user_id as the group name…

One approach might be to have an aggregate store of user-related information, either on a centralised node or distributed using something like Riak Core or Mnesia, and have all nodes write to and read from there.

A simpler option, though potentially a bottleneck, would be the following: on each node run a regular named GenServer that dispatches to the local user processes through Registry. You can then use multi_call and abcast to broadcast a tuple {uid, message} to that named process, and each node would unwrap the message and relay it locally using the via-tuple in uid. For calls you may have to spawn a process to handle the reply asynchronously, to avoid blocking the dispatcher.

1 Like

I think that it’s possible to use any term as the group name, as is usually
the case with systems like these.

Yes, both syn and swarm can register arbitrary erlang terms as groups / names. I heartily recommend syn, by the way. It’s very pleasant to work with. I’ve never used it for anything big, though, but it seems to do a fairly decent job of auto-synchronizing names/groups on/after netsplits.

Thanks a lot for the replies, guys! I will give syn/swarm a try, if you say that it can accept arbitrary terms as group names then I’m happy :slight_smile: