I’ve been using Passenger Standalone instead of WEBrick or Mongrel for a while and it’s great, but Pow simplifies things even further: now I don’t even have to restart the server when switching between projects.

The only thing that bothered me is that it installs a firewall rule to forward traffic from port 80 to 20559. I do some PHP work under the Apache that ships with OS X and I don’t want to add a port number to my URL there, nor do I want to uglify those nice new .dev URLs for Pow.

Oddly enough, though, this isn’t true for me. If I go to localhost, up came the expected page from Apache, while foo.dev went to Pow. I didn’t specify a port for either.

What’s happening is that localhost is resolving to an IPv6 address (::1) and the ipfw rule doesn’t cover IPv6, only IPv4.

In this case, “it’s not a bug, it’s a feature” is 100% true. I hope Pow never “fixes” this.

Update June 29: Sadly, while I haven’t updated Pow, Mac OS X 10.6.8 now resolves “localhost” to the IPv4 address and this little trick has stopped working for me. Using the IPv6 loopback address is still OK: http://[::1]/.

Major enhancements include:

• Automatic creation of regular expressions specific to CSS and Javascript for use in <style> and <script> tags, as well as external files marked with appropriate content types. This avoids false matches, in particular for Javascript when a private server’s entire URL space is mapped to a subdirectory on the public server.
• Support for remapping URLs in style attributes.
• A new flag allows a URL mapping to be ignored for styles.
• Fixed a bug that silently discarded HTML comments when ProxyHTMLExtended was on. IE conditional comments look like HTML comments, and so were lost prior to this fix.
• HTML comments are scanned for URL matches, again to fix up resources referenced inside an IE conditional comment.
• URLs are always mapped in the order they are defined. Global mappings (outside of a <Location> directive) are always applied first.
• Fixed the HTML 4.01 DOCTYPEs so that they always emit a complete URI.
• ProxyHTMLExtended is more fine-grained and can be enabled for just styles or scripts, as well as both (the old behavior when set to “on”).

The development of this module was funded by the Council of Better Business Bureaus in late 2008. Some internal discussion held up the release of the code until now, but I’m pleased it is being made available to the Apache community. Note that because my work on this module took place two years ago, I can offer only very limited support for it.

The source code is available on GitHub.

I’m using httperf to simulate client load on the test server and quickly decided that --wsesslog looked like the best choice for simulating an actual browser’s effect on the server.

A problem: how to generate those session workloads? I certainly don’t want to do this by hand for even one page. I want to generate a hit on every file referenced by the target page, but ignore anything hosted elsewhere.

A solution:

#!/usr/bin/env ruby
 
require 'rubygems'
require 'hpricot'
require 'open-uri'
 
if ARGV.length < 1
  $stderr.puts "usage: #{$0} url
 
  'url' must include the protocol prefix, e.g. http://"
  exit 1
end
 
url = ARGV.shift
if url =~ %r{^(https?://)([-a-z0-9.]+(:\d+)?)(.*/)([^/]*)$}i
  $protocol = $1
  $host = $2
  $document_dir = $4
  document_url = $5
else
  $stderr.puts 'Could not parse protocol and host from URL'
  exit 1
end
 
doc = Hpricot(open(url))
 
def puts_link(uri)
  return if uri.nil?
 
  if uri =~ %r{^#{$protocol}#{$host}(.*)$}
    puts "    #{$1}"
  elsif uri !~ %r{^https?://}
    if uri =~ %r{^/}
      puts "    #{uri}"
    else
      puts "    #{$document_dir}#{uri}"
    end
  end
end
 
puts "# httperf wsesslog for #{url} generated #{Time.now}"
puts
 
puts "#{$document_dir}#{document_url}"
 
(doc/"link[@rel='stylesheet']").each do |stylesheet|
  puts_link stylesheet.attributes['href']
end
 
(doc/"style").each do |style|
  style.inner_html.scan(/@import\s+(['"])([^\1]+)\1;/).each do |match|
    puts_link match[1]
  end
end
 
(doc/"script").each do |script|
  puts_link script.attributes['src']
end
 
(doc/"img").each do |img|
  puts_link img.attributes['src']
end