Idapted TechTeam

ajax_load improvement

guolei — Wed, 08 Sep 2010 08:24:30 +0000

Currently the ajax_load method is kind of messy. If there’s multiple ajax_load in one page, one can only send request when the previous request is finished. And there’s always one js error “doRequest_xx is not defined.”

This is caused by the bad design of the first version. When I had lunch with @Blade this noon, we got some new ideas. Here’s the refactored code.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72

def ajax_load(url, options={})
@ajax_counter = (@ajax_counter || 0) + 1
index = options[:index] || @ajax_counter
dom_id = Idp::Util.uuid
c = "
"
c << %{<div id="#{dom_id}" class="#{options[:class]}">#{options[:show_loading]==false ? "" : idp_image_tag("loading.gif")}

}
c
end

After refactor ,all the requests are send at one time. And as there’s limited number of rails process, if you’d like one request to be sent earlier, you can identify “index”, like

1	<%= ajax_load(url, :index => 0) %>

The smaller the index is, the sooner it will be sent.

Client HTTP Caching in Rails

blade — Mon, 06 Sep 2010 08:48:36 +0000

HTTP Cache is a browser Cache.
HTTP is typically used for distributed information systems, where performance can be improved by the use of response caches.
This kind of cache used for reduce request times and bandwidth.
more info.

Let’s talk about its principle.

The Browser start a request. When request reach Web Server. the web server will check the request head.
if request head have description about cache-control and data have not modified. Web Server will return 304.
the browser receive 304 will render the result last request.

In sales project. A student detail page use *ajax_load* split as multipart requests.
I think every single ajax_load request can use HTTP Cache.

Before HTTP Cache Optimaztion:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

Firebug:
GET http://www.veecue.com/sales/students/60329/calls 200 OK 245ms

Log:
Processing Student::CallsController#index (for 127.0.0.1 at 2010-09-06 16:35:13) [GET]
Parameters: {"student_id"=>"60329"}
Student Load (1.3ms) SELECT * FROM `students` WHERE (`students`.`id` = '60329') LIMIT 1
Call Load (71.5ms) SELECT updated_at FROM `calls` WHERE (`calls`.student_id = 60329) ORDER BY updated_at desc LIMIT 1
Call Load (1.2ms) SELECT * FROM `calls` WHERE (`calls`.student_id = 60329) ORDER BY created_at desc
SaleStatus Load (0.8ms) SELECT * FROM `v2_system_production`.`list_items` WHERE (`v2_system_production`.`list_items`.`id` IN (327,329,320))
SaleStatus Load (0.8ms) SELECT * FROM `v2_system_production`.`list_items` WHERE (`v2_system_production`.`list_items`.`id` IN (326,329))
EqFollowType Load (0.8ms) SELECT * FROM `v2_system_production`.`list_items` WHERE (`v2_system_production`.`list_items`.`id` IN (99,104))
EqCallResult Load (0.6ms) SELECT * FROM `v2_system_production`.`list_items` WHERE (`v2_system_production`.`list_items`.`id` IN (337,4))
Rendering student/calls/index
Rendered student/calls/_list (9.1ms)
Completed in 245ms (View: 170, DB: 74) | 200 OK [http://localhost/students/120490/calls]

Use HTTP Cache

1
2
3
4
5
6
7

class Student::CallsController < Student::BaseController
def index
if stale?(:last_modified => @student.calls.first(:order=>"updated_at desc", :select => "updated_at").updated_at.utc, :etag => "call")
@calls = @student.calls.all(:order => "created_at desc", :include => [:status_before, :status_after, :eq_follow_type, :eq_call_result])
end
end
end

Restart server.
Then open student detail page again.
The first time return 200.

When I refresh.It return 304.

1
2
3
4
5
6
7
8

Firebug:
GET http://localhost:3000/students/120490/calls 304 Not Modified 26m
Log:
Processing Student::CallsController#index (for 127.0.0.1 at 2010-09-06 16:38:33) [GET]
Parameters: {"student_id"=>"120490"}
Student Load (1.4ms) SELECT * FROM `students` WHERE (`students`.`id` = '120490') LIMIT 1
Call Load (0.8ms) SELECT updated_at FROM `calls` WHERE (`calls`.student_id = 120490) ORDER BY updated_at desc LIMIT 1
Completed in 8ms (View: 0, DB: 2) | 304 Not Modified [http://localhost/students/120490/calls]

The speed improve a lot. cool,

Summary:
1.suit for a page not often modified.
2.the are 3 ways to control HTTP Cache. need further learning.
3.only GET request can use HTTP Cache.
4.may be it has way for ajax use HTTP Cache.

development tips on Titanium

Andy — Wed, 01 Sep 2010 06:51:43 +0000

After releasing our first iPhone app, I just learned a lot about using Titanium SDK. Today I want to share some valuable development tips, which are mostly gathered from Titanium developer’s forum.

1. Remember that the device is not as speedy as the simulator.
2. When you need to use a webView, try to communicate with the app as less as possible. If you need to create new elements in the interface, do it before you launch the webView and not while the webView is active, and attach it to the window from the webView. If you need to make calculations ( for example to catch the touchmove coordinates) inside webView, do it in the javascript from inside he webView, and send the result to the app at the end ( on touchend maybe ).
3. If you have to store persistent data, but is not an sql data ( for example you need to remember some settings the user is able to configure ) use Titanium.App.Properties and not Titanium.Database. Properties are persistent so you may want to ‘remove’ them in your app.js – to start afresh.
4. Test your application on the device( if possible ). You will find that you might need to improve the usability – for example you might ned to add a “Loading…” view to mask a slow loading component, and provide in the same time feedback to the user. ( this is somehow related to the point 1. )
5. It takes minimal extra effort to target the Android platform as well as the iPhone. Take a bit of extra time to make sure your app works on both (using platform detection where you need to do something slightly differently, or need to skip a function that causes the app to crash on one device).

if (Titanium.Platform.name == 'android') {
    // Android stuff
}

Connect to test data-sources by detecting whether the app is running in a simulator.

if (Titanium.Platform.model == 'google_sdk' || Titanium.Platform.model == 'Simulator') {
    // Simulator variables
}

6. Ti.API.info the name of the JS file at the start, so you can capture the sequence of navigation when testing. Also useful if you could have obsolete files to check they are not used.
7. About HTTPClient:
[1] Titanium HTTPClient runs in a separate thread and often (always?) doesnt release it. Use global client to download multiple files to minimize number of threads. You can make a recursive procedure, that gets called from onreadystatechange or onload (if it fires)
[2] HTTPClient.onload often doesnt fire, use onreadystatechange with readyState == 4
[3] Titanium HTTPClient doesnt close itself automatically, so before reopening it with a different direction, use “abort”, or it will bring you the same file as before.
[4] So if u create a client, all your subsequent events and timers will fire in several threads. Titanium threads seem to share variables, so if your event is sensitive to doubled execution, use a simple integer or boolean to create a semaphore.

Dynamic add class method in (class

fred — Wed, 01 Sep 2010 06:26:22 +0000

1	@person = Person.new

As you know, @person is a instance object of Person, in the other words @personal.class is Person. but do you know who is Person.class? sure, it’s Class, so we can say Person is a Class’s “instance object”. Next, we add some class methods to Person by difference ways.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

class Person

def self.class_method1
end

class << self
def class_method2
end
end

end

def Person.class_method3
end

class << Person
def class_method4
end
end

........

class Class
def class_methodX
end
end

Wait, does the last solution work? yes, not only we could call Person.class_methodX, but also call in any class. for example

1
2
3
4
5
6
7
8
9
10
11

class Class
def whoami
puts 'you are Class'
end
end

class Me
end

Me.whoami
# => you are Class

The last solution is bad because it effects too many classes, at the same time it can’t dynamic add class method in runtime, not only itself, all above solutions can’t. Therefore, I write a method like below

1
2
3
4
5
6
7
8
9

class Person
def self.create_method(name, &block)
self.send(:define_method, name, &block)
end
end

Person.create_method(:hello) {puts 'hello'}
Person.hello
# => undefined method `hello' for Person:Class (NoMethodError)

Why it doesn’t work? After a seconds, I got it.

1 2	Person.new.hello # => hello

Look, we just add a method on “self”, then the “self” is Person, so the method is instance method instead of class method. But how we can add a class method dynamically? Shall we add method to Class? No, we should add it to a stuff who it like Class but only for Person. Yes, the “class << self; self; end" can do this. Change above code like this

1
2
3
4
5
6
7
8
9

class Person
def self.create_method(name, &block)
(class << self; self; end).send(:define_method, name, &block)
end
end

Person.create_method(:hello) {puts 'hello'}
Person.hello
# => hello

Above is my experience to research how to use (class << self; self; end). Besides, I got some informations about it: (class << self; self; end) is a virtual class and it can't be instance

1
2
3
4
5
6
7
8
9
10
11
12

class Person
def self.virtual_class
class << self; self; end
end
end

v = Person.virtual_class
p v #=> #
p v.class #=> Class
p Class.new # => #
v.new #=> can't create instance of virtual class (TypeError)
p Class.new.new # => #<#:0x10039f840>

Summary
In above example, (class << self; self; end) is a kind of Class, it as same as Class but only for Person.

How to analyze and forecast?

denny — Wed, 01 Sep 2010 03:20:31 +0000

What we can get from history data? and how we can get it? These are two question when I face to the LPD function. and In trainer side , there are many statistics data. can we get something from these data? The answer is sure, we can get , but how?

These two data I think over it and get some idea, so share with us for discussing.

Analyze Data

For LPD, we want to know how much time our almost learner spent on one page and scenario ? how many point they got ? we want to get the standard time_spent and point for every page or scenario, so we can apply right study method to our learner. Till now, we have the data about this. so we need analyze it and find some result.

From the aspect of Math, we can induce this to solve the function .

μ : 正态分布的位置参数，描述正态分布的集中趋势位置。正态分布以X=μ为对称轴，左右完全对称。正态分布的均数、中位数、众数相同，均等于μ

σ : 描述正态分布资料数据分布的离散程度，σ越大，数据分布越分散，σ越小，数据分布越集中。也称为是正态分布的形状参数，σ越大，曲线越扁平，反之，σ越小，曲线越瘦高。

we want to get the value of μ for every page or scenario in time spent. so for this , we can draw one chart for this, and get the below value for analyzing.

Mean (µ): sum of all data divided by the number of data points; aka the “average” Example: {4, 8, 10} µ = (4+8+10)/3 = 7.33
Median: the middle value of a set of data Example: {4, 8, 10} med = 8
Quartile: quarter of the data; ‘Q1’ and ‘Q3’ are the values at the 25th and 75th percentiles; (Q2 = med = 50%) Example: {2, 4, 8, 9, 10, 13} Q1 = 4; med (Q2) = (8+9)/2 = 8.5; Q3 = 10
Standard Deviation (σ): measure of how spread out the data is or how close your data is to the mean Example: µ = 0; σ = 1

http://www.idapted.com/lpr/lpd/pages/84357?item=time_spent

Forecast Data

For this task , it is so big and difficult that there is no good way for this, but we can start from some easy and simple case .

1. we want to know how many shift should be opened at pm platform , for this, we should first to know how many student will be valid in that time and forecast how many student will take a class.

2. we want to know the relationship between student who take the class and the interaction number

commit_shift_num(y) : the shift number of trainer signed during pm platform time

actual_student_num(x1) : the number of student who take a class during pm platform time

valid_student_num(x2) : the number of student who can take a class during pm platform time

history_num(x3) : the number of history_record which duration great than two mins

How to get this data, there are some statistics methods for this .

一元线性回归预测分析法( 1)
二元线性回归预测分析法(2)
时间序列预测法(3)

Object : y = a + b*x1

Step :

1. get the number of x2 , this should use the 时间序列预测法,

2. get the relationship between x1 and x2, this should be x1 = a + b*x2

3. get the y = a + b*x1

Code :

1. the chart for flow (4), for this, i need more work, till now , it not finish, but we can get the trend about the number of valid student , it is up.

2. the code for 一元线性回归预测分析法

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

def line_alg(cond,x,y)
puts "*".center(58, "-")
puts "Sample : #{cond}"
n = Sample.count(:conditions=>cond)
x_ave = Sample.average(x.to_sym,:conditions=>cond)
y_ave = Sample.average(y.to_sym,:conditions=>cond)
sxx = 0
syy = 0
sxy = 0
Sample.all(:conditions=>cond).each do |s|
yi = s.send(y) #commit_shift_num
xi = s.send(x) #actual_stu_num
sxx += (xi-x_ave)**2
sxy += (xi-x_ave)*(yi-y_ave)
syy += (yi-y_ave)**2
end
b = sxy/sxx
a = y_ave - b*x_ave
puts "sample number : #{n}"
puts "y_ave : #{y_ave}"
puts "x_ave : #{x_ave}"
puts "y = #{a.to_f.round(2)} + #{b.to_f.round(2)}*x"
r = sxy/Math.sqrt(sxx*syy)
puts "r = #{r.to_f.round(2)}"

end

Result :

1. the relation between actual_student_number and valid_student_number

1
2
3
4
5
6
7
8

y : valid_stu_num
x : actual_stu_num
Sample : valid_stu_num > 0 and actual_stu_num >0
sample number : 225
y_ave : 108.3289
x_ave : 45.5733
y = 24.64 + 1.84*x
r = 0.3

From the result, we can get the relationship is loose, r <0.3, but I think the reason is the sample is so small and some of them is not right, so when we have large sample and correct data, we can get the function for this.
2. the relation between actual_student_number and history_number

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

y : history_num
x : actual_stu_num
----------------------------*-----------------------------
Sample : history_num > 0 and actual_stu_num >0
sample number : 1268
y_ave : 43.6538
x_ave : 22.0039
y = 5.55 + 1.73*x
r = 0.97
----------------------------*-----------------------------
Sample : history_num > 0 and actual_stu_num >0 and sample_date >= '2010-07-01' and sample_date < '2010-08-01'
sample number : 31
y_ave : 83.129
x_ave : 48.3548
y = 1.52 + 1.69*x
r = 0.88
----------------------------*-----------------------------
Sample : history_num > 0 and actual_stu_num >0 and sample_date >= '2009-01-01' and sample_date < '2010-01-01'
sample number : 352
y_ave : 72.0881
x_ave : 35.8665
y = 14.17 + 1.61*x
r = 0.92
----------------------------*-----------------------------
Sample : history_num > 0 and actual_stu_num >0 and sample_date >= '2009-10-01' and sample_date < '2010-08-01'
sample number : 295
y_ave : 76.4949
x_ave : 43.539
y = 6.38 + 1.61*x
r = 0.91

As you can seen, all the r greater than 0.9, so we can deduce the relation between these two item is very close.

3. relationship between actual_student_number and commit_shift_number

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

y : commit_shift_num
x : actual_stu_num
----------------------------*-----------------------------
Sample : commit_shift_num > 0 and actual_stu_num >0
sample number : 1254
y_ave : 29.6858
x_ave : 21.7568
y = 9.85 + 0.91*x
r = 0.89
----------------------------*-----------------------------
Sample : commit_shift_num > 0 and actual_stu_num >0 and sample_date >= '2010-07-01' and sample_date < '2010-08-01'
sample number : 31
y_ave : 48.6129
x_ave : 48.3548
y = 37.78 + 0.22*x
r = 0.33
----------------------------*-----------------------------
Sample : commit_shift_num > 0 and actual_stu_num >0 and sample_date >= '2009-01-01' and sample_date < '2010-01-01'
sample number : 352
y_ave : 46.4631
x_ave : 35.8665
y = 23.27 + 0.65*x
r = 0.66
----------------------------*-----------------------------
Sample : commit_shift_num > 0 and actual_stu_num >0 and sample_date >= '2009-10-01' and sample_date < '2010-08-01'
sample number : 295
y_ave : 51.4441
x_ave : 43.539
y = 37.13 + 0.33*x
r = 0.44

all the r great than 0.3, even the sample number is 31, so we should build the relationship between these tow item, and get the factor for y = a +b*x, if we have this function, we can forecast the shift number be opened.

First iPhone App : IELTS 31 Formuals

Andy — Sat, 28 Aug 2010 07:08:49 +0000

In last two month, our team worked for Idapted first iPhone App. We spent a lot time on releasing this app as Apple’s process is really awful to a first user. Well, everything was done in the last. Now our first iPhone App is available in Apple Store. It’s free in current version. you can install it from here: http://goo.gl/ht4p. It’s based on Titanium 1.4. so a big thanks to the brilliant Appcelerator Team.

We open sourced this app on our github repos. We also built a website for this app at http://www.31formulas.com

Enjoy it and let us know if you have any feedback on it.

Apple iPhone app : Debug and Release through Titanium

Andy — Wed, 25 Aug 2010 04:16:09 +0000

Once people develop an iPhone App using Titanium1.4, they must be very happy when the App is running on iPhone simulator. They feel great until they decide to debug their app on their devices or even release their app to AppleAppStore. This happens specially to these who firstly do this. Why? There are many reasons that you will hate Apple during the debug or release process. I bat you are like me then.

Assume that: You’ve already finished coding an iPhone app in Titanium 1.4 framework. Now you want to install your app on your device. If you do this first time, guess what will happen to you? Let me tell you my story.

You will be blocked by Titanium certificates detections when you want to install your app to your device. see below image:

Well, in order to become an iPhone developer, you should pay Apple for an iPhone Program Permission. My boss paid Apple for an iPhone Program Team permission, it cost him about one month to pass Apple’s approval. After that, he, as an admin agent iPhone developer, invited me as an iPhone developer member with admin permission. Here we had been running into issue for two days. Finally, we found that he only can invite my exist Apple ID as an iPhone developer, because I can’t finish a registration process after he invite me with a non-exist Apple ID. The evil is Apple indicates that, we should invite non-exist accounts as team members.

Now have I got the permission to debug my app on my iphone? Wait a moment. Actually I should be invited twice separately, one is for iTunesconnect. another one is for Provisioning Portal(Dev Center). These two sites are different, only in Dev Center, I can config my provisioning profile, which is a main concept for releasing iPhone App or allowing to install App on certain devices. But in iTunesconnect, I need to release my app to iTunes/Apple App Store.

So far, I’ve got enough permission to debug the App on my own iPhone. But I am still unable to distribute my App to others’ devices. How that happens? As an iPhone developer, I need to pass Titanium’s detection of several certificates. only after that, I can install app on my iPhone. (PS: if you want to test your app on Android, the only need is to connect your device.)

In following images, I will show you the steps through which you can pass all the Titanium detection of certificates.

Then you will get the following files after your above configuration.

After you import the above certificates to your keychain, you should see something similar in your Keychian.

Finally, your Titanium will be the “ready status” for your debug and release on real device.

The whole process might be painful unless you understand every concepts that Apple creates. I encounter problems when creating distribution provisioning profile. Only admin agent developer has permission to do so. Thus, you should ask admin agent for this provisioning profile, he can share this profile with a private key file, you import this key file to your Keychian, then you can distribute the app to whatever devices which is bundled into specific provisioning profile.

rails small tips

blade — Wed, 04 Aug 2010 07:45:33 +0000

1. flash and flash.now

flash in rails used to pass messages.it is use session do this.

1
2
3
4
5
6
7
8
9

def create
if User.authorize(params[:login], params[:password])
flash[:notice] = "login success"
redirect_to target_path
else
flash[:error] = "login fail"
render :new
end
end

flash[:notice].
it effect after redirect_to.
in this example if login failed the error message will not show in template “new” but show in next redirect page.

when we use render should use *flash.now*.
flash.now[:notice] will effect in current page.

2.select and include

Rails ‘s ActiveRecord finders ‘s “select” option do not work when “include” preload assoic records。

condition:

1
2
3
4
5
6
7

class Student < ActiveRecord::Base
has_one :call
end

class Call < ActiveRecord::Base
belongs_to :student
end

how to show student’s call’s updated_at.

before optimize

1	@students = Student.all(:include => [:call], :select=> "students.id, students.login, students.called_times, calls.updated_at as last_called_at")

it do work, but in sql statment, It still preload “call” and return a full record.

after optimize

1
2
3

#include eql left outer join
@students = Student.all(:select => "students.id, students.login, students.called_times,
calls.updated_at as last_called_at", :joins => "left outer join calls on students.id = calls.student_id")

now, table students only select the fields we need.

calculate results VS. store result sets

Andy — Wed, 04 Aug 2010 03:01:43 +0000

Firstly, what’s the results in title? For example, the results may be learners who will be alive in coming two weeks, or the results may be study times of one learner, etc. Generally, results should be some data coming from analyzing other data.

Here my question is, how can we get certain results much faster and easily?

Case 1:
We want to pick up a name list, all people in the list should be under certain condition, say, condition A, condition B, condition C, so normally, we just select the results which obey the rule(condition). Right? It depends.

If name list is always dynamically different in condition A,B,C, it’s really hard to store the results set, we have to calculate instantly to make sure the results is accurate.

If name list is very difficult to figure out in condition A,B,C, we should store the stable part of the result set, so that we will not get stuck when calculate tons of data eevery time.

Case 2.
We want to show study times of a learner, how can we do that? Should we store the study times as one attributor of study record or should we just make one study record as one-time study? Obviously, it depends on the complication of study record and the definition of one-time study.

In our case, one study record is instantly changed when it’s in different status, so storing study times in study record is not a good idea because the results might be locked when status changes. But if we calculate study times based on its status, it will be more reliable and accurate. and also it will not be locked when status is with exception.

Conclusion:
To store result set, is a way to cache results, but some times it’s not easy to maintain the result sets, for instance, if there are too many conditions when getting a results set, it’s always in heavy load to update result sets, you have to listen to every condition level changes. If the situation is complex, remember to cache your result set and maintain it carefully.

To calculate results instantly lets you ignore the real data, you just need to use the rule carefully, that will be more accurate, because you every time select what the rule defines. However the disadvantage is, when the data sets is large, you have to pay more to do that instantly, so you want to do a cron job or just cache the stable part of previous result sets.

Deploy dozens of rails applications into one domain

seven — Tue, 03 Aug 2010 02:35:07 +0000

We eventually have more than 30 rails applications running. Back to the days when we had only 3, we deploy each to a separate domains:

1
2
3

app1.example.com
app2.example.com
app3.example.com

They are beautiful. But, the 3 apps actually working together to do one big task. It would looks better to be served in one domain:

1
2
3

www.example.com/app1/
www.example.com/app2/
www.example.com/app3/

The advantage to use one domain is that it’s more beautiful – You may don’t even notice it’s actually 3 apps! And now they finally grew to more than 30s by using the great framework and architecture – eco_apps (http://github.com/idapted/eco_apps).

The other advantage more than beautiful is that you will never worry about the javascript cross-domain requests, direct or AJAX.

Perhaps the easiest way to deploy multiple apps to subdirs is to use passenger:

http://www.modrails.com/documentation/Users%20guide%20Nginx.html#deploying_rails_to_sub_uri

http://www.modrails.com/documentation/Users%20guide%20Apache.html#deploying_rails_to_sub_uri

But we had bad experience with passenger. One rails app stuck on some memcache server failure. Passenger was trying to start it over and over again and blocked all processes – the whole server suddenly stopped responding to anything! Perhaps they already fixed that, but we still think it’s better to serve rails “out” of Nginx or Apache.

First we need all rails apps to run in a subdir, no matter mongrel

1	mongrel_rails --prefix=/app1 ...

unicorn:

1	unicorn_rails --path /app1

or swiftiplied/evented mongrel or thin.

We use Unicorn(http://github.com/blog/517-unicorn) for rails and Nginx as a reverse proxy:

Nginx conf:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38

upstream app1{
server localhost:10010;
}

upstream app2{
server localhost:10020;
}

upstream app3{
server localhost:10030;
}

...

upstream app30{
server localhost:10300;
}

server {
listen 80;
server_name www.example.com;

location / {
root /home/app/root;
}

location /app1 {
root /home/app/app1/current/public;
proxy_pass http://app1;
}

location /app2 {
root /home/app/app2/current/public;
proxy_pass http://app2;
}
.....
}

That’s it. Take a look at http://github.com/idapted/eco_apps to see how we make dozens of rails apps working together.