[Tabularium-devel] Fwd: Run Tornado servers on all available CPUs with pre-forking

[CaStarCo <cas...@gm...>]

Un mensaje de la lista de tornado :) , en cuanto liberen la nueva
versión cambiaré el código y quedará simplificado :D , simplemente
detectará el número de procesadores de forma automática y creará un
proceso independiente por cada uno de ellos ^^. Además compartirán
puertos, así que que nginx no tendrá que perder tiempo balanceando
carga y de ello se encargará el kernel.


---------- Forwarded message ----------
From: Bret Taylor <bt...@gm...>
Date: 2009/12/9
Subject: Run Tornado servers on all available CPUs with pre-forking
To: pyt...@go...


I just checked in a (experimental and backwards-compatible) change to
enable "pre-forking" Tornado servers, and I would love your feedback:
http://github.com/facebook/tornado/commit/6fb90ae694190fcedc48d9fb98b02325826d783e
Previously, every Tornado server ran on a single thread within a
single process. If your server had 8 cores, you would have to run 8
separate Tornado processes on 8 different ports to utilize all CPUs
and put a reverse proxy like nginx in front to balance requests
between all those processes (see
http://www.tornadoweb.org/documentation#running-tornado-in-production).
With this change, you can easily run a single Tornado server listening
to a single port on all available CPUs of a server.
To try out this change, change your call from listen(port) to
bind(port) followed by start(), e.g.,
    http_server = tornado.httpserver.HTTPServer(application)
    http_server.bind(options.port)
    http_server.start() # Pre-forks multiple child processes
    tornado.ioloop.IOLoop.instance().start()
start() detects the number of CPUs on the current machine and
"pre-forks" that number of child processes so that we have one Tornado
process per CPU, all with their own IOLoop. You can also pass in the
specific number of child processes you want to run with if you want to
override this auto-detection. listen(port) is simply a shortcut for
bind(port); start(1).
Background
The "pre-forking" phrase is used to mean a couple of different things,
so it is worth clarifying what it means for Tornado. Essentially, we
bind a single server socket to a port, and we fork() one process per
CPU. Each of those CPUs calls accept() on the shared server socket,
and the Linux kernel gives new requests to one of the child processes
on a first-come-first-serve basis. Each of the child processes has
their own epoll-based IO loop, and a single request is handled
entirely within one of the child processes. There is no shared memory
or shared state between the forked child processes, only a shared
port.
This technique is used by a number of high performance servers,
including Unicorn (see http://tomayko.com/writings/unicorn-is-unix).
Apache pre-forking means something entirely different (forking a
process for each request), so please don't get confused if you Google
"pre-forking" and see people talking about the performance
characteristics of Apache.
Bret



-- 
- Per la llibertat del coneixement -
- Per la llibertat de la ment...       -