Using environment vars in lighttpd config files

Our game is hosted on a farm of servers behind a load balancer. All servers are identical except for their names (prod-01, prod-02, etc) and virtual IP addresses.

In PHP, if I try to access $_SERVER['SERVER_NAME'], I only got the domain name "prod.game.com". Actually, this was exactly the same var as the $_SERVER['HTTP_HOST'].

For logging purposes, I needed to know the name of the server that my script was currently running on. So I updated my lighttpd.conf

Lighty has a feature called include_shell that you can use in its config files. It will basically run a shell script and add its output to your file.

So I wrote a simple shell script to define a var.serverName (this is a custom value, name it as you want, but keep the var prefix) and then re-use when needed.

#!/bin/bash
echo 'var.serverName="'$(uname -n)'"'

Then, I included it in my lighttpd.conf file using include_shell "/etc/lighttpd/scripts/serverName.sh"

To define the PHP SERVER_NAME value :

setenv.add-environment = (
  "SERVER_NAME" => var.serverName
)

To add it as a Server: response Header :

server.tag = var.serverName

Note that include_shell directives are only called when you start lighty and not on every request.

Incorrect MySQL date

Several ingame time calculation we made in game are based on Paris time. Some weeks ago, we decided to make a pass on all our servers to always use Paris Time (GMT +1).

Today, I spotted that logs we save in the DB have some date inaccuracy. It appears that our mysql server and instances weren't always updated to the correct date. Some hours later, here is what I learned :

Finding and updating MySQL date

You can tell what timezone mysql should use when you start the service. If you don't specify anything, it will use the system time. Once loaded, you can get its time by running SELECT NOW().

This is the easiest way to spot errors.

To know the defined timezone, run SELECT @@global.time_zone. If not defined, you'll read SYSTEM, which is not very helpful.

Note here that even if you changed the system date AFTER you started mysql, mysql will still use the date that was in effect when you first launched it.

It means that changing your server time will not affet running mysql processes. You'll have to restart mysql to do that : sudo /etc/init.d/mysql restart

Finding and updating the server date

Even after restarting mysql on some servers, the mysql date was still incorrect. After connecting the the sql server, I found that it was the server time that was incorrectly set (I just type date).

To update the current time zone, I had to call sudo dpkg-reconfigure tzdata (I'm using ubuntu) and choose the correct city

Updating mysql running through ndb_mgm

I'm no server expert, so this part was a little trickier. Some of our databases are using ndb cluster for replication. Reloading those configurations was harder.

First, I had to connect to the server running the ndb management and call ndb_mgm. In the later prompt, I typed show and this get me the list of all servers currently managed.

I then shut them down typing shutdown.

The, I reloaded the management and the node running on this server by doing sudo /etc/init.d/mysql-ndb-mgm restart and sudo /etc/init.d/mysql-ndb restart

Finally, I had to connect to all the servers I saw earlier (with the show command) and run sudo /etc/init.d/mysql-ndb restarton each of them

ParisJS #8

(The whole conferences were in french, but as I'm writing every post on my blog in english, I'll write this one too.)

Tonight I was attenting the 8th parisjs meetup. We had the chance to have it hosted at Valtech, in a very wealthy part of Paris, in somptuous quarters and a large room with giant screen (and great pizzas too).

Anyway, this time the meetup was split in two : first part was made of complete talks, while second part only had small lightning talks. With pizza and beer in between.

Wakanda

The first talk was about Wakanda. Wakanda is a complete suite of client/server Javascript which goal is to help create business oriented apps. It is created by the same company as 4D, but this time is completly open source.

The product seemed well done, but kinda monolithic : server side js, client side javascript framework, application studio, debugguer.

I can surely see it used for creating Intranet applications, for some very specific business cases. I have an ambiguous feeling about Wakanda. I think they did a very good job, the whole package seems really neat, the IDE is clear, client/server communication is smooth, etc. On the other hand I'm always suspicious of apps that can do everything.

(Note that I didn't try the product, I only heard about it tonight for the first time and watched a 20min presentation so that's a little light for me to give a precise opinion)

Joshfire Framework

A few months ago, @sylvinus from Joshfire told us that they planned to realease their framework, modeslty named Joshfire.

This is by far the most interesting presentation of the meetup. Joshfire is a javascript framework aimed at working seamlessly between any devices : desktop, mobile, iPad, TV, anything.

It's basic idea is to abstract to its bare bone any interaction. From what I remember, for any "page" you want to define you have to write two JSON structures.

One defines the data tree (what your data is), the other defines the UI tree (how your data can be interacted with). It also ships with some adapters (an adapter is the link between the UI Tree and a specific device).

All you have to do to port your Joshfire app to a new device is to write (or grab if it already exists) an adapter. The main logic and data code is the same for every device, the maintenance is now only focused on any specific device.

Even if the main idea of Joshfire is awesome, their documentation and licensing is even more awesome : the project is already completly documented, with examples and is also completly free and available on github.

CSON, no sorry, JISON

@jie was supposed to talk about CSON, a compressed JSON. But as he already gave a very similar talk (about BSON) at the last

parisnode event, he changed the subject.

I'm not really sure what the real subject of the talk was. I think he mentionned compiling Javascript into Javascript (much like CoffeeScript), but it eventually turned into a Jison demo.

Jison seems to be a tool for automating lexing and parsing languages, to output new code. In effect, this could allow for transforming a custom language into an existing language. CoffeeScript, Haml or Sass works that way : they take a new language as input, and return Js, Html or Css as output.

The advantages are that the custom language is usually much more concise, easy to write and easy to read.

Anyway, this presentation was full of nice quotes :

The language I'm about to show you is totally useless.

Pipe ? What is that ? Oh, you mean the strokes.

The bad thing about Jison is that its documentation is crap

Why does it work ? It was supposed to fail !

Overview of a dev life

This talk wasn't about any new shiny technology. It was about those old browsers and bad habits we had 10 years ago, and how we still managed to make great websites.

This remind me of the "good ol' days" when I discovered the web with FrontPage and ClaraSoft. When IE6 was considered a modern browser.

@molokoloco show us that he managed to simulate ajax using nested iframe, had a website still online and running even after 10 years, was able to code full websites in 3 days, and had read more than 50.000 blog posts. And all that while still using Windows and Dreamweaver.

(Small side note : I praise both @molokoloco and @jie for what they managed to do. I know how easy it is to procrastinate and configure UI, test new tools and IDE instead of getting things done. That didn't seem to be an issue for them.)

mongoDB

This talk was the first in the serie of lightning talks (less than 20mn). It was about mongoDB and how to integrate it in a nodejs app.

As I had never tried nor nodejs nor mongoDB I didn't quite grasp what it was really about. All I get is that in order to get a result from a mongoDB query, one have to go through 3 steps of connection to the DB. In the asynchronous world of nodejs, this doesn't work well, so each step has to fire a callback once executed.

From there, I was lost. There was something about manually calling a callback once initialized, but honestly I didn't get it.

Anyway, I learned from a small talk during pizza time that mongoDB was better than couchDB (performance-wise), and that one have to sacrifice binding to the altar of redundancy to get really fast response time.

Audio demo in less than 1kB. No, 800B. No, wait, 200B

This time the talk was a very fun demo. Using some simple minification techniques, Js was able to play a random music.

This involved encoding the wav file in base64 (taking special care of some chars), and then playing it in an audio tag. Adding a bit of random to play a different note on each loop.

Fun, but interesting.

Some words about the Observer Pattern

This talk was more about theory than practice. Maybe it was the late hour, but it didn't catch my attention as the other did. Maybe someone would re-explain it to me ?

Well, I got the general idea : coding a js snippet that hard code the name of the elements that should react is bad practice because of the maintanability nightmare it cause when elements are removed.

Instead, the speaker advocate the use of a MessageBroker. Some kind of general listener that then dispatch event to element registered to that event (am I right ?).

Then someone in the audience noted that the opposite pattern should work better : that every element registered itself to an event and react when such event is fired (am I still right ?)

Honestly, I really am not sure about this talk, so if anyone could fill the blanks and correct me, I'd be glad.

Sencha and Sencha Touch

I won't go into too much detail on that talk. This was a generic presentation of Sencha (formerly known as ExtJs). I already had the basic understanding of ExtJs (but not the expertise of some of the ExtJs experts in the room), so I was really only half listening.

Sencha is a great product. It is very similar to what Wakanda was showing in the first talk. It is destined to data-heavy applications with lots of numbers, columns, charts, pies, etc.

Sencha Touch is a mobile version of Sencha. It is iOS oriented and emulate the standard UI elements in HTML.

Mozilla Labs

A quick list of the Mozilla Labs projects. I honestly don't remember them all, except Tilt that is a 3D viewer of any web page. It simply calculate the DOM depth and create a pyramid-like structure that you can browser in 3D.

I'm sure that beyond the cool and shiny aspect, it can be useful as a debug/optimisation tool.

Conclusion

Once again, thanks to #parisjs for those talks, always inspiring and a way to meet new people.

For all the new readers, I have an unfixed bug in my comment section : your post can be invalidated if your are using Chrome with an email auto-filler. My custom anti-spam will flag you as spam because Chrome fills an hidden honeypot field.

Working on Wednesday #4 : Rails for Zombies

(I finally managed to update Ubuntu to the 10.10 version, configure Unity to make it more usable, and I'm not ready to start the RoR tutorial I wanted to start last week.)

I used vim this week (instead of nano) for very simple file editing. I will force myself to use only vim on wednesday to really learn it. I plan to completly drop Komodo Edit once I'll be familar enough with vim to gain more time than I lose.

Rails for Zombies

Anyway. I finally get to start the Rails for Zombies tutorial. And here is what I learned.

There are two distinct things I'm going to learn here : Ruby and Rails. I discovered that I was already familiar with a lot of basic concepts of Rails, thanks to cakePHP. Model and table convention, model relationships (belongs to, has many, etc) were exactly the same.

Ruby on the other hand was new. But also very interesting. No need to add () after a method to call it when no args are passed, possibility to chain method of the same object (much like jQuery) and the use of dynamic (yet strongly typed) vars.

The first part of the tutorial (basic CRUD) was easy. I didn't find how to make a complex SQL query like this (cake) one :

$Zombie->find('all', array(
'order' => array('Zombie.graveyard' => 'ASC', 'Zombie.name' => 'DESC'
));

I guess it's just a matter of time before I learn how to to do in Rails anyway. I saw that there was more "advanced" method (much like the cake find method) that would surely allow such finding.

Part 2 : Validation and Relationships

The second part of the tutorial was more about models. Validation rules were pretty much the same as those of cakePHP. But the Ruby syntax makes it more concise.

I couldn't find how to retrieve the validation errors when saving an object using the create method (no problem when manually calling save, though).

I enjoyed being able to create binded instances by simply adding the binded element to the model. Not very clear isn't it ? Here some code to explain it better. The last two lines have exactly the same effect :

class Zombie < ActiveRecord::Base
end
class Weapon < ActiveRecord::Base
belongs_to :zombie
end
Weapon.create(:name => "Rocket Launcher", :zombie_id => 2)
Weapon.create(:name => "Rocket Launcher", :zombie => Zombie.find(2))

Later on, I had to select all weapons binded to a specific Zombie. I was able to find the a specific zombie using any of the two following commands :

z = Zombie.where(:name => "Ash").first
z = Zombie.find(:first, :conditions => { :name => "Ash" })

Then, I wanted to find the weapons binded to this zombie. First I tried the following command :

Weapon.where(:zombie_id => z.id)

Obviously, this worked. But I wanted to test some more of the Rails magic. So I tried something along the lines of what I did in the previous chapter and tried to use my z var directly instead of z.id.

Weapon.where(:zombie => z)

This didn't worked. Well, I guess I'll also learn why later on. But one even more weird thing is that the following command did work even I'm not really sur why ?

Weapon.where(:zombie_id => z)

That's all I did today (well I also did some more Mercurial and Dingux testing, but that wasn't really work)

Alternate content with SWFObject

SWFObject allow a web developer to add an alternate content in place of the swf file is the user doesn't have flash installed.

In the case of a web app this is great to provide your user with a link to the Adobe website. You then tell your users Flash is needed and provide a link for them to install it.

Unfortunatly, if for one reason or another the swf file you try to load is unreachable (maybe your host is down ?), SWFObject will display the same alternate content. Meaning that a user with a perfectly correct Flash version will see an error message telling him that he does not have Flash when in fact the error is on your side.

That's not very user friendly and makes you look bad.

So, to fix this, I added two error messages as alternate content :

<div id="alternateContent" class="alternateContent">
  <div class="noFlash message">It seems that you do not have the Flash player installed. Please install it, by <a href="http://www.adobe.com/go/getflashplayer" target="_parent">following this link</a>.</div>
  <div class="error404 message">Sorry, we were unable to load the game. Please try again in a few moments.</div>
</div>

Then, in CSS I decided to hide them both :

.alternateContent .message { display:none; }
.alternateContent.noFlash .noFlash { display:block; }
.alternateContent.error404 .error404 { display:block; }

And finally, in Javascript I checked for the current Flash version. If it's equal to zero, it means that Flash is not installed, so I display the noFlasherror message, otherwise, I guess it's a 404 error and display the other message.

var flashVersion = swfobject.getFlashPlayerVersion(),
  alternateContent = $('#alternateContent')
;
// Displaying one message or the other
if (flashVersion.minor=='0') {
  alternateContent.addClass('noFlash');
} else {
  alternateContent.addClass('error404');
}

This is not bulletproof : I only test for two cases. And a better solution would also have been to put the error message dynamically using Javascript instead of polluting the HTML markup with contradictory text.