Is `channel_id` keep in it the node ref?

Phoenix Forum Questions / Help
,
1

Phoenix:
I’m using Phoenix as a Broker, different clients (Android and iOS connected to PhoenixWeb.UserChannel) and multiple instances of Microservices (MS), connected to PhoenixWeb.ServerChannel, in Go. all will be connected to multiple instances of Phoenix (inter-connected via Node).

Presence:
I do use Presence to keep all information of Client and MS connected to Phoenix

Client Presence:

%{
      node: Node.self(),
      channel: socket.topic,
      channel_pid: socket.channel_pid,
      user_id: socket.assigns.user_id,
      venue_id: socket.assigns.venue_id,
      device_id: socket.assigns.device_id,
      room_type: socket.assigns.room_type,
    }

MS Presence:

%{
      node: Node.self(),
      channel: socket.topic,
      channel_pid: socket.channel_pid,
      user_id: socket.assigns.user_id,
      group: socket.assigns.group,
      name: socket.assigns.name,
    }

Send Message:
From ServerChannel, when an MS send a message to a client, first, it will get the list of Client Present and filter it to get the right item. Then, we will do this call below, which will call a method into PhoenixWeb.UserChannel and do the push. All good this work perfect :D…

send(item[:channel_pid], {:send_msg, %{"topic" => topic, "playload" => playload}})

Question:
Case: I do have N1 and N2 (instance of Phoenix). Client 1 is connected to N1 and MS is connected to N2. MS send message to Client 1

  • I did try and it’s working but I’m wondering how Phoenix know the channel_pid is from N1 or N2?
  • What will happen, if I have the same PID name on N1 and N2 but each is linked to different process? Phoenix will still send the message to the right Node?
2

The general suggestion is to avoid sending messages to channel pids. Instead, have channels join a relevant topic, and then broadcast to that topic. Your questions seem to indicate that you think that Phoenix does channel pid to channel pid messaging across the cluster, but that isn’t how it works. Channels subscribe to a topic via a local node registry. Individual nodes also connect to one another via one of several backends (pg2, redis, etc).

When a process broadcasts, it does two things (simplified to ignore fastlane):

  1. uses the local node registry to send to every pid subscribed to the topic
  2. also broadcasts to every remote node. Each remote node then does step 1.

To be clear, if the nodes are connected via distributed erlang you can just do send(pid) without caring whether pid is on your node or a remote node. Erlang is just cool that way. However, maintaining a distributed registry would be a serious challenge, so Phoenix opts for the aforementioned two stage approach. This also has the perk of lowering inter-node bandwidth since if there are 100 pids on the remote node you don’t need to send the same data 100 times.

4 Likes
3

There are use-cases though where sending directly to a channel_pid can be ok, which is when you know you want to send a msg to just one connection/user. I’ve worked on device pairing, where server and device communicate with each other with that last week. Another example is PhoenixLiveView, which also basically operates in a 1to1 messaging mode.

1 Like
4

@benwilson512 and @LostKobrakai.
Thanks for you feedback,
as @LostKobrakai said, yes I need to send a message to one specific device at some point, and not broadcast to all of them.
So, from you said, it it ok do to it but not recommended if it to send the same message for multiple users.

5

It depends on what you mean by a user. Keep in mind that if the same person has 3 tabs open, that is 3 different channel pids. If you want a notification to appear for that person, you really want to handle that via a topic broadcast and not via pid messaging. There’s no guarantee those 3 channel pids are even all on the same node!

2 Likes
6

Yes, I got this part. PID will be used only if I want to a single socket user/device. A topic broadcast on the other hand will be used for a notify a group of sockets, like user/devices, users/devices, etc.

@benwilson512, I have another little question: If I do call Process.alive?(channel_pid), I will be able to know if the pid is still alive on my local node or a remote node? Base on what you said, if the nodes are connected via distributed erlang you can just do send(pid) without caring whether pid is on your node or a remote node, am I right?

7

Yes, Process.alive? works on all pids, regardless of what node they’re on, as does send and all of the genserver functions.

3 Likes