Questions[long-polling](server)

Normally when you pull the network cable out of a network device (which simulates some drop on the network, so an obscure disconnect), the underlaying sockets do not detect a disconnect (at least our units don't) so we use keep alive to drop connection and reconnect.

All units use hourly long poll to nginx server. So my question is:

If the unit does not detect a cable disconnect (and probably 100 other situations on the network), then also nginx does not disconnect the socket and that socket stays open?

Stay open for how long?

Should we use on the nginx side also tcp keep alive towards client? How?

I have an ASP.NET application that uses some long polling magic.

The problem comes when the server has accepted some requests and it's just waiting to return them. If I restart apache in that moment I get multiple instances of mod-mono-server4.exe that together use 100% of the CPU. It stays in that state until you restart the Ubuntu instance.

I can't replicate it consistently, it happens 1 in any 10-20 attempts and it happens in production too.

Does anybody know what the problem may be?

I'm trying to design an architecture to handle massive TCP socket connections and I'm in doubt on the limits of this architecture.

I'll have to handle ~20k concurrent TCP connections and these are long-polling connections, they'll remain connected for long periods of time and will be sending data every minute.

Using threads is out of question, since 20k threads will starve the system resources. I'm planning to use gevent to handle such amount of concurrent connections or even use haproxy and 2 servers (w/ gevent) to handle 10k connections each for instance. Does that makes sense ? Does anyone have some advice or had some experience using gevent with 10K+ connections ? Does anyone have idea of the hardware requirements to handle these connections ? I saw some benchmarks that shows lots of connection timeouts for gevent on 5k concurrent connections, what is not very promising for my problem.

Note: I've already read about the C10k problem and the advices on the "Million-user Comet Application".

Also posted here: https://stackoverflow.com/questions/8839295/long-polling-options-nginx-php-node-js

I'm designing a long-polling app to broadcast small changes very rapidly to, possibly, a large number of users. The app will run in tandem with a website running a fairly standard cms. They'll both be running on one server, and to begin with so will the database.

I come very much from a LAMP environment and I'm definitely a developer and not a sys-admin. That said I'm not afraid to try out some new things.

I've spent the day researching my options and I'm hoping people can answer some questions and give me some recommendations.

I have narrowed it down to these:

A. Apache and php for the website, Node.js for the app (running on a separate ip)
B. Nginx and php for both the website and app
C. Nginx and php for website, Nginx and Node.js for the app

So the questions:

1. How does Nginx handle PHP's non-multi-threadingness ;). Will PHP prove as much a bottleneck as Apache would have for long polling?
2. I've heard it suggested that I use nginx as a reverse proxy in front of Apache and Node.js, is that a better solution than just nginx? If so, why?
3. Which option have you used/would recommend?

Bear in mind that ease of set-up could be a factor, I'm fairly comfortable with Apache but I've only played with Node (I have strong javascript) and I've never installed Nginx.

I'll happily provide clarifications if anyone needs them.

I use nginx in this mode for BOSH and chat clients along with gzip.

location ~* /http-bind/ {       
     proxy_buffering off;
     keepalive_timeout 55;
     access_log off;
     tcp_nodelay on;
     proxy_pass http://x.x.x.x:1111;
}

Is this the best approach to managing long polling in nginx.

I also use just one worker process for altogether for web & chat (single CPU). Is that fine?