Inconsistent browser retry behaviour for timed out

I am experiencing occasional retries for a POST request, when there is no response from server due to timeout. All modern browsers have retry logic for idempotent requests (GET, HEAD, etc) but I am unable to reason out why it happens for a POST request.

I am testing this case using a simple node.js server with 3 routes and chrome browser .

/       : gives a html page with jquery and code snippets to fire ajax requests
/hi     : gives a text response 'hello'
/sleep  : request will timeout without any response

By default, node.js http server times out a request after 2 minutes.

retry.js

var http = require('http');
var server = http.createServer();

server.on('request', function(req, res) {
    console.log(new Date() + ' ' + req.method + ' request on ' + req.url);

    if (req.url === '/sleep') {
        console.log('!!! sleeping');
    } else if (req.url === '/') {
        html = "$.post('/hi', {'for':'server'}, function() { console.log(arguments) } ).error(function() { console.log(arguments) })";
        html += "<br><br>";
        html += "$.post('/sleep', {'for':'infinite'}, function() { console.log(arguments) } ).error(function() { console.log(arguments) })";
        html += '<script src="http://ajax.googleapis.com/ajax/libs/jquery/1.9.1/jquery.min.js"></script>';

        res.writeHead(200, {'Content-Type': 'text/html'});
        res.end(html);
    } else {
        res.writeHead(200, {'Content-Type': 'text/plain'});
        res.end('hello');
    }
});

server.listen(2020);
console.log('server listening on port 2020');

run it

$ node retry.js 
server listening on port 2020

1

Load this page in browser http://localhost:2020

Fri Mar 01 2013 12:21:59 GMT+0530 (IST) GET request on /
Fri Mar 01 2013 12:21:59 GMT+0530 (IST) GET request on /favicon.ico

2

From Dev console, fire an ajax POST request to /hi using jquery

$.post('/hi', {'for':'server'}, function() { console.log(arguments) } ).error(function() { console.log(arguments) })

Fri Mar 01 2013 12:22:05 GMT+0530 (IST) POST request on /hi

3

Fire a POST request to /sleep, results in a retry after 2 mins and errors out after 4 mins.

$.post('/sleep', {'for':'infinite'}, function() { console.log(arguments) } ).error(function() { console.log(arguments) })

server logs shows 2 requests

Fri Mar 01 2013 12:22:21 GMT+0530 (IST) POST request on /sleep
!!! sleeping
Fri Mar 01 2013 12:24:21 GMT+0530 (IST) POST request on /sleep
!!! sleeping

Firing it again, errors out in 2 mins without any retry.

Fri Mar 01 2013 12:30:01 GMT+0530 (IST) POST request on /sleep
!!! sleeping ?

It's not getting retried until we fire a request to /hi (or any other url) that results in a response. And retry happens for just one subsequent request to /sleep.

In browser, the network tab shows the pattern like

/hi - success 
/sleep - cancelled - 4 mins (retry happens)
/sleep - cancelled - 2 mins (no retry)
/sleep - cancelled - 2 mins (no retry)
/hi - success 
/sleep - cancelled - 4 mins (retry happens)
/sleep - cancelled - 2 mins (no retry)
/sleep - cancelled - 2 mins (no retry)

Question

Though we need to design our web app to tolerate these extra requests (either by browsers or any other intermediaries), this inconsistent browser retries looks weird. I had observed this behaviour in chrome (v16 & v24) & firefox.

Can someone help me to understand the browser retry logic behind timed out non-idempotent requests ?