You get the onclose event. Reconnection can't be done automatically by SockJS, so it's your responsibility to handle that accordingly.

Then how should you handle the onclose event to reconnect? I don't see anything in the API docs like a 'connect' method... so would you have to create a new SockJS instance? In that case, how do you clean up all the event listeners you made for the old SockJS instance?

Thanks

so would you have to create a new SockJS instance?

yes. it's similar to websockets API

In that case, how do you clean up all the event listeners you made for the old SockJS instance?

The naive assumption is that once you stop referring to the old sockjs instance it should be garbage collected. If that's not true - file a bug as this would be a memory leak!

I do just that with onclose, here's a sample of my javascript

$(function() {
    var chat, sockjs, offline_members;
    var sockjs_url = '/chat/sjs/';
    var reconnect = true;

    new_conn = function() {    
        sockjs = new SockJS(sockjs_url);

        sockjs.onmessage = function(e) {
            // chat(JSON.stringify(e.data));
            chat(e.data);
            // chat(e);
    };
        sockjs.onclose = function(e) {
            if (reconnect){ new_conn();}
        };

This is perfectly fine. I would suggest some kind of delay and exponential backoff - you don't want to spin 100% cpu if the network is unavailable, right?

yea, it's on my todo already to set up exponential backoff, one of the first things I plan on doing after I finish my backend rewrite to use tornado.gen where possible.

I've commited a simple reconnector at https://github.com/knowitnothing/sockjs_reconnect