sockem

Warning: sockem is rated YMMV. At present, this repo doesn't have its own automated tests or many users.

What?

Sockem is a library that wraps the ActionCable client to "dumb it down" so that it can be used for three things:

1. sockem.subscribe() to a single channel (SockemChannel) and automatically re-subscribe if the connection is lost for as long as the app needs it
2. Allow the user to sockem.request(payloadObj, cb) with some assurance that the request will be retried until a matching response is received and cb(responseObj) called
3. sockem.unsubscribe() when the channel is no longer needed

How?

$ npm install sockem

import * as sockem from 'sockem'

sockem.subscribe()

sockem.request({id: 42}, (res) => {
  document.write(res.name)
  sockem.unsubscribe()
})

And, to make that do something, a Ruby class in app/channels/sockem_channel.rb that looks something like:

class SockemChannel < ApplicationCable::Channel
  def subscribed
    stream_for current_user
  end

  def handle_answer(data)
    response = {
      name: data["id"] == 42 ? "Pants" : "Not Pants",
      request_id: data["requestId"]
    }

    self.class.broadcast_to(current_user, response)
  end
end

You'll want to set up the above current_user with some scheme that matches up the HTTP session with the channel (see: connection.rb in the example app)

Why?

Sockem is an extraction from KameSame.com that may help others who want to use ActionCable for use cases that don't resemble chat rooms or push notifications. One example might be an app that can't use HTTP2 but needs to send a lot of very small payload requests with as little latency as possible, because they block the user experience. To be more specific, this code is used by KameSame's client JS to send users' flashcard answers to the server and wait (and retry) until all the outstanding requests get a response, reconnecting as often as necessary.

Warnings

Other than lacking a test suite or many users, keep note that this library's retry mechanism can really go wild if your server ever starts spitting errors, as every affected client will essentially hammer it. If anyone's interested, adding a more sophisticated retry mechanism (like a backoff) would be an improvement. So would a retry limit.