A channel == a process

I think there’s a minimum of 2 processes for any given channel. There will be one PubSub process, and one process for each client/user.

When you create a channel like that it creates no processes.

However let me go through the whole process as I understand it (from debugging and reading a good bit of the code).

The client websocket/longpolling connects to the ‘Socket’ that you made, that spawns off a new process, thus (1) process so far, for the open socket. That socket process then checks if it is allowed via your own code and such, do whatever you want, set assigns that will be given to all other topic processes too. That socket then checks the topic being connected to and spawns another new process for just that topic (and each new topic connected comes in the same socket process but then makes a new topic etc…). That topic process gets a copy of the socket assigns (this way you can do something like get the user from the database and store in an assigns for all topics to see). So at this point you have two processes, and one additional process for each extra topic.

Thanks

If I understood it correctly, then a topic like “user:*” will be spawned once for all users, each user/client has be connected over a persistent socket (which itself is a process, sure), but there will be no additional processes for each user:#{id} topic, right?

Nope, a process is spawned for each topic for each user. You can join more topics from inside a topic though, and ‘those’ will not create more processes, but anytime the javascript sides connects to a topic for every single user then a new process is created for that specific topic/connection by default.

Hmm, thank you, so., this can potentially become expensive.. I am wondering, how we could send a single message to an opened socket connection, on a single users topic but to an explicit single user. I mean simply like responding is happening..

I mean, you mentioned, that its possible to join an user to a topic without creating those processes.

So basically from JS channel = s.join('users') Inside users one would join to user:id but broadcasting messages wouldn’t reach the client side.. because there is no channel with the topic user:id defined..

If you broadcast to a topic unique to a given user, say "user:123", only that user will receive the message. If you’re asking how to message to a single user 123’s devices, then you can use Phoenix Presence, but we need to know more info to know what your exact usecase is. Generally it’s just fine to broadcast to a private user topic since you usually want all tabs/devices the user has joined to receive the message.

Not too bad at all. Process’s are overall cheap.

You need to make at least one topic, as it is ‘that’ that can then join others without extra processes. You can broadcast just fine once the topic you made registers to them, like a socket joins "users" but that topic also internally joins "user:#{id}", then you could broadcast to "user:#{id}" and that "users" topic would get it just fine.

Hi @chrismccord thank you for your attention. I’ll try to explain the use case:

10 users, each on a single device. All of them connect are connecting to the backend via WebSockets.
(For each of those is a socket process created, so far so good)

Now all of them join the room:lobby Another process is created, which holds the state of the channel and all its meta data.

Now each of them joins to the user:uuid channel.. As I understood, this will spawn another 10 processes (one for each wildcard uuid)

So, basically, the main question is, how send a single message to a particular user, (I could store a map like uuid -> socket on some generic topic like users ?

To a particular user? Just MyEndpoint.broadcast("user:uuid", ...) as normal.
What about that does not work for you?