From: Eric Wong <normalperson@...>
Date: 2010-07-31T17:32:12+09:00
Subject: [ruby-core:31559] non-blocking I/O constantly clearing cache in 1.9.2

Hi all,

The addition of IO::WaitReadable and IO::WaitWritable modules being
extended into Errno::EAGAIN / Errno::EINPROGRESS exceptions means the
method cache gets cleared every time those exceptions are raised.  From
my experience with non-blocking I/O, EAGAIN is a very common "error" and
can be raised many times a second.  Thus the method cache gets cleared
constantly.

While I don't write/maintain large Ruby applications with many methods,
in them, I doubt the method cache would exist in Ruby if it didn't help
performance, so constantly clearing it probably hurts some applications
that run under 1.9.2.

I don't yet know of any big apps doing non-blocking I/O (using Ruby IO
methods like IO#write_nonblock/IO#read_nonblock) with lots of traffic
running 1.9.2.  If there are, it would be interesting to see if
(and maybe how much) the following monkey patches help performance:

	class Errno::EINPROGRESS
	  include IO::WaitWritable
	end

	class Errno::EAGAIN
	  include IO::WaitWritable
	  include IO::WaitReadable
	end

Personally, I've never been a fan of caches.  Yes, I use caches and even
occasionally introduce them, but I still think they're ugly, last resort
hacks.  If the monkey patches above don't help performance a significant
amount, maybe we could get rid of the method cache entirely :)

-- 
Eric Wong