Errors going to 2 dynos on Heroku with socket.io / socket.io-redis / rediscloud / node.js

Question

I have a node.js / socket.io app running on Heroku. I am using socket.io-redis with RedisCloud to allow users who connect to different dynos to communicate, as described here .

From my app.js:

var express = require('express'),
app = express(),
http = require('http'),
server = http.createServer(app),
io = require('socket.io').listen(server),
redis = require('redis'),
ioredis = require('socket.io-redis'),
url = require('url'),
redisURL = url.parse(process.env.REDISCLOUD_URL),

And later in app.js ...

var sub1 = redis.createClient(redisURL.port, redisURL.hostname, {
    no_ready_check: true,
    return_buffers: true
});
sub1.auth(redisURL.auth.split(":")[1]);
var pub1 = redis.createClient(redisURL.port, redisURL.hostname, {
    no_ready_check: true,
    return_buffers: true
});
pub1.auth(redisURL.auth.split(":")[1]);
var redisOptions = {
    pubClient: pub1,
    subClient: sub1,
    host: redisURL.hostname,
    port: redisURL.port
};
if (io.adapter) {
    io.adapter(ioredis(redisOptions));
    console.log("mylog: io.adapter found");
}

It is kind of working -- communication is succeeding between dynos.

Three issues that happen with 2 dynos but not with 1 dyno:

1) There is a login prompt which comes up and works reliably with 1 dyno but is hit-and-miss with 2 dynos -- may not come up and may not work if it does come up. It is (or should be) triggered by the io.sockets.on('connection') event.

2) I'm seeing a lot of disconnects in the server log.

3) Also lots of errors in the client console on Chrome, for example:

socket.io.js:5039 WebSocket connection to 'ws://example.mydomain.com/socket.io/?EIO=3&transport=websocket&sid=F8babuJrLI6AYdXZAAAI' failed: Error during WebSocket handshake: Unexpected response code: 503

socket.io.js:2739 POST http://example.mydomain.com/socket.io/?EIO=3&transport=polling&t=1419624845433-63&sid=dkFE9mUbvKfl_fiPAAAJ net::ERR_INCOMPLETE_CHUNKED_ENCODING

socket.io.js:2739 GET http://example.mydomain.com/socket.io/?EIO=3&transport=polling&t=1419624842679-54&sid=Og2ZhJtreOG0wnt8AAAQ 400 (Bad Request)

socket.io.js:3318 WebSocket connection to 'ws://example.mydomain.com/socket.io/?EIO=3&transport=websocket&sid=ITYEPePvxQgs0tcDAAAM' failed: WebSocket is closed before the connection is established.

Any thoughts or suggestions would be welcome.

Answer 1

Yes, like generalhenry said, the issue is that Socket.io requires sticky sessions (meaning that requests from a given user always go to the same dyno), and Heroku doesn't support that.

(It works with 1 dyno because when there's only 1 then all requests go to it.)

https://github.com/Automattic/engine.io/issues/261 has a lot more good info, apparently web sockets don't really require sticky sessions, but long-polling does. It also mentions a couple of potential work-arounds:

• Roll back to socket.io version 0.9.17, which tries websockets first
• Only use SSL connections which, makes websockets more reliable (because ISP's and corporate proxies and whatnot can't tinker with the connection as easily.)

You might get the best results from combining both of those.

• You could also spin up your own load balancer that adds sticky session support, but by that point, you're fighting against Heroku and might be better off on a different host.

RE: your other question about the Node.js cluster module: it wouldn't really help here. It's for using up all of the available CPU cores on a single server/dyno,