Phoenix websocket processes getting reused? Stray pubsub message

I have a custom websocket process that occasionally receives a pubsub message it never subscribed to. It’s not a very complex module, and the pubsub message is for an entirely different customer, so I’m fairly confident that I’m not subscribing in my websocket module. I was pretty sure each websocket connection got its own process, but I’m not quite sure where else to look. The pubsub message received was most likely intended for a live view process, which points me towards the process maybe being reused.

I add the socket to my endpoint like this:

  socket "/my-resource/:id/socket", MyAppWeb.MySocket,
    websocket: [path: "/", connect_info: [:uri, :x_headers]],
    longpoll: false

MyAppWeb.MySocket implements Phoenix.Socket.Transport. One thing that caught my attention was the section on building custom transports:

Whenever the transport receives a new connection, it should invoke the connect/1 callback with a map of metadata. Different sockets may require different metadata.

If the connection is accepted, the transport can move the connection to another process, if so desires, or keep using the same process. The process responsible for managing the socket should then call init/1.

I am using the connect callback to start a linked process, but the messages it sends back are entirely different from the pubsub ones I’m seeing, and it seems like the default websocket transport invokes connect/1 and init/1 in the same process.

Any thoughts on what might be causing this? I’m using phoenix 1.18.4 and bandit 1.10.3