Blaine Cook on Scalability

In this very interesting interview with Blaine Cook (once of Twitter fame) Blaine provides the best analogy for describing “performance” vs. “scalability” I’ve yet to read:

Performance and scalability are very different things. Performance is like the speed limit; it’s a question of how quickly you can get from point A to point B. In this case, we’re talking about how quickly you can get a page rendered and delivered to the user.

Scalability is a question of how many users you can handle. Highways intrinsically don’t scale, for example, because when they’re full, you get traffic jams. In web architectures, what we aim to provide are systems that will expand (usually by adding more hardware) to handle more traffic.

Obviously they’re related – if you have a traffic jam, then the effective speed limit is lower than the theoretical limit. But increasing the speed limit won’t make traffic jams any better.

There are a whole bunch of ways to make traffic less congested – adding more lanes to the highway, encouraging people to use public transit, or better yet encouraging people to work closer to home.

Likewise, there are many techniques for making web sites more scalable, and most of them don’t involve making things much “faster”.

As an aside, I continue to be impressed by Blaine’s grace under the seemingly relentless “Tw/ails doesn’t scale and it’s your fault” onslaught. Good on you sir.

Unread Messages Count in LovdByLess

Recently Less Everything released their “social network in a box” platform Lovd by Less, built upon the Rails framework, and it’s pretty nice. Very easy to set up and get running. I’m enjoying hacking around in it.

One of the first things I noticed was that users weren’t told how many unread messages they have when in their dashboard. Easy enough to add, just make these two small changes:

1. In _private.html.erb change the Messages link to:

<%= link_to( "Messages (#{@p.unread_messages.length.to_s})", messages_path ) unless @p.received_messages.empty? && !@p.has_network? %>

2. In profile.rb change the has_many association to:

has_many :unread_messages,   :class_name => 'Message', :conditions => ["messages.read=?",false], :foreign_key => 'receiver_id'

(I suspect the change above is actually a fix to an existing bug since calling @p.unread_messages as-is throws a SQL error).

Jump-start on Rails 2.0: Rails2 PeepCode

Ruby on Rails 2.0 was released on Dec 6, 2007 and fortunately for me I had the luck of starting a brand new Rails project on Dec 9. A brand new slate on which to try out all the new Rails toys.

Invaluable in this has been Ryan Daigle’s Rails2 PDF, a well-written compendium of what’s new, what’s changed and examples thereof. It also contains the complete Rails changelog, which seemed silly to me at first glance but has since been surprisingly useful as a desktop, keyboard-side reference in its own right.

Admittedly all the information contained within can probably be found on the web and piece-meal if you manage to read all the major Rails blogs each day but for $9 having it summarized like this has made it well worth the price as a bridging document until the Rails 2.0 books start to appear.

Rails Functionality Into Plugins

With the release of Rails 2.0 quite a bit of core functionality has been pushed out into plugins, presumably to lighten the core and allow for these plugins to take development paths and release schedules independent of that of Rails itself.

I just ran head-first into this while trying to use in_place_edit_for which generated this error:

undefined method `in_place_edit_for’ for FaqQuestionController:Class

The docs don’t make it immediately obvious (or rather, they’re completely obtuse) about which plugin the functionality has been moved to. Fortunately one can figure this out for one’s self via the Svn repository:

· CHANGELOG
· account_location/
· acts_as_list/
· acts_as_nested_set/
· acts_as_tree/
· atom_feed_helper/
· auto_complete/
· continuous_builder/
· deadlock_retry/
· exception_notification/
· http_authentication/
· in_place_editing/
· javascript_test/
· legacy/
· localization/
· open_id_authentication/
· scaffolding/
· scriptaculous_slider/
· ssl_requirement/
· token_generator/
· tzinfo_timezone/
· tztime/
· upload_progress/

(Thanks to Ryan Bigg via the rubyonrails-talk list.)

Don’t Preface Partials With Numbers

I just learned today that it’s a bad idea to name your Rails partials starting with a number, ie: _4_happy_places.html.erb. Turns out Rails throws the following error if you do:

`@4_happy_places’ is not allowed as an instance variable name

This has been tested against Rails 2.0.1.

Google Map Markers in YM4R/GM

When using YM4R/GM, by Guilhem Vellut, in your Rails app chances are you’ll want to mark points on the map using custom icons. The means to do so is not readily apparent or straight-forward. I’ve found that this works well:

map = GMap.new( Constants::GM_DIV_ID, Constants::GM_NAME )

# Define the start and end icons
map.icon_global_init( GIcon.new( :image => "/images/google_maps/icong.png", :icon_size => GSize.new( 24,38 ), :icon_anchor => GPoint.new(12,38), :info_window_anchor => GPoint.new(9,2) ), "icon_start")
map.icon_global_init( GIcon.new( :image => "/images/google_maps/iconr.png", :icon_size => GSize.new( 24,38 ), :icon_anchor => GPoint.new(12,38), :info_window_anchor => GPoint.new(9,2) ), "icon_stop" )

icon_start = Variable.new("icon_start"); icon_stop = Variable.new("icon_stop")

if respond_to?( "start_lat" ) && respond_to?( "start_long" )
    map.center_zoom_init( [start_lat, start_long], Constants::GM_ZOOM )
    map.overlay_init( GMarker.new( [start_lat, start_long], { :icon => icon_start, :title => name + " start", :info_window => "start" } ) )
end

if respond_to?( "end_lat" ) && respond_to?( "end_long" )
   map.overlay_init( GMarker.new( [end_lat, end_long], { :icon => icon_stop, :title => name + " end", :info_window => "end" } ) )
end

Reloading dispatch.fcgi

Rails development is usually done using dispatch.cgi so changes are readily apparent. However dispatch.cgi is painfully slow for a production environment so it only makes sense to use dispatch.fcgi. You can check which you’re using by looking into your /public/.htaccess file. In this case I’m running dispatch.cgi and .fcgi is commented out:

RewriteRule ^$ index.html [QSA]
RewriteRule ^([^.]+)$ $1.html [QSA]
RewriteCond %{REQUEST_FILENAME} !-f
RewriteRule ^(.*)$ dispatch.cgi [QSA,L]
#RewriteRule ^(.*)$ dispatch.fcgi [QSA,L]

Problem is dispatch.fcgi won’t immediately recognize changes made to content. In order to force a flush of all dispatch.fcgi instances we could kill all the processes but this seems better to me:

[chris@enki ticklist]$ touch public/dispatch.fcgi