499s and repeated requests

So, I’m running a rather standard nginx → unicorn → rails application,
over HTTPS. Recently, I started looking into some strange load bursts
have been going on for a while but are getting more important), and saw
repeated requests (both GET and POST) in my access logs, all getting a
status code and reporting 0 bytes sent. I know the browser isn’t seeing
these 499s, but that they’re closing the connection before getting data.
These aren’t long-running requests; I can find them in my Rails logs and
they’re returning in 50-100ms.

I’ve seen in other threads that “the user clicked save twice really
fast” is
the most likely cause – but we had one client make 6500 requests over
course of a few minutes the other day, so, um, it’s probably not that.
rather sure this isn’t some kind of denial-of-service attack, either;
appear to be legit requests repeated by… the browser…? insanely
I can’t, alas, find any way to replicate this locally.

I know there are application-level things we can do to mitigate the
of something like this (duplicate records being created and such), but I
want to check at this end: is there some kind of configuration error I
have made that would cause this?

In case it matters, we appear to still be running nginx 1.1.19.


Posted at Nginx Forum:

On 07.11.2013 04:35, njvack wrote:

they appear to be legit requests repeated by… the browser…? insanely

I experienced that kind of issue in either Firefox or Chrome (i think it
was Firefox) myself. The only difference is that the status code was
444. After the closing of the connection by nginx Firefox would try to
repeat the request, instead of showing an error message, until I closed
the browser window or nginx served a valid page.

The nginx I was running at that time was an nginx 1.4.3 built by dotdeb.