How to structure a service that has many variation? - ruby-on-rails

I got my app ready, which rely on Ruby on Rails as backend.
Now I am going to publish another app, which has similar functionality, but different topic. like StackExchange, it has multiple sites under this big umbrella, like Stackoverflow, Superuser, Game Development etc.
I have these few approaches in mind:
Same code base, deploy to multiple Rails apps.
Same code base, handle by same Rails app, but with flagging. to identify.
How do Stackoverflow handle this kind of variation?

It will be much easier for you, initially, to have a single app handling the different sites, and to use the url to set some settings for each site, such as which data to load up, which stylesheets etc. In this way you're designing a single rails app, and can test it easily with various localhost urls which trigger the different variants etc.
If your site becomes very successful you will want to split it across multiple servers anyway. At that point, you will need to consider a strategy for doing so: you may decide that you will have one server per site, or it might be the case that one of the sites is so massively successful that you need multiple servers for that and one to handle the others, or something else. That's a nice problem to have, and you might not have it, so keep it simple for now is my advice.
EDIT - for info on how stackexchange works, go to http://blog.serverfault.com/ . Searching for "architecture" could be a good start for you.

Related

What tools to use for a website with lots of "realtime" page updates (coming from a Rails background)?

We are planning to make a "large" website for I'd say 5000 up to many more users. We think of putting in lots of real time functionality, where data changes instantly propagate to all connected clients. New frameworks like Meteor and DerbyJS look really promising for this kind of stuff.
Now, I wonder if it is possible to do typical backend stuff like sending (bulk) emails, cleaning up the database, generating pdfs, etc. with those new frameworks. And in a way that is productive and doesn't suck. I also wonder how difficult it is to create complex forms with them. I got used to the convenient Rails view helpers and Ruby gems to handle those kind of things.
Meteor and DerbyJS are both quite new, so I do expect lots of functionality will be added in the near future. However, I also wonder if it might be a good idea to combine those frameworks with a "traditional" Rails app, that serves up certain complex pages which do not need realtime updates. And/or with a Rails or Sinatra app that provides an API to do the heavy backend processing. Those Rails apps could then access the same databases then the Meteor/DerbyJS app. Anyone thinks this is a good idea? Or rather not? Why?
It would be nice if anyone with sufficient experience with those new "single page app realtime" frameworks could comment on this. Where are they heading towards? Will they be able to handle "complete" web apps with authentication and backend processing? Will it be as productive/convenient to program with them as with Rails? Well, I guess no one can know that for sure yet ;-) Well, any thoughts, guesses and ideas are welcome!
For things like sending bulk emails and generating PDFs, Derby let's you simply use normal Node.js modules. npm now has over 10,000 packages, so there are packages for most things you might want to do on the server. Derby doesn't control your server, and it works on top of any normal Express server. You should probably stick with Node.js code as much as possible and not use Rails along with Derby. That is not to say that you can't send messages to a separate Rails app, but since you already have to have a Node.js app running to host Derby, you might as well use it for stuff like this.
To communicate with such server-side code, you can use Derby's model events. We are still exploring how this kind of code works and we don't have a lot of examples, but it is something that we will have a clear story around. We are building an app ourselves that communicates with an email server, so we should have some real experience with this pretty soon.
You can also just use a normal AJAX request or send a message over Socket.IO manually if you don't want to use the Derby model to do this kind of communication. You are free to make your own server-side only routes with Express along with your Derby app routes. We think it is nice to have this kind of flexibility in case there are any use cases that we didn't properly anticipate with the framework.
As far as creating forms goes, Derby has a very powerful templating system, and I am working on making it a lot better still. We are working on a new UI components feature that will make it possible to build libraries of self-contained UI widgets that can simply be dropped into a Derby app while still playing nicely with automatic view-model bindings and data syncing. Once this feature is completed, I think form component libraries will be written rather quickly.
We do expect to include all of the features needed for a normal app, much like Rails does. It won't look like Rails or work like Rails, but it will be similarly feature complete eventually.
For backend tasks (such as sending emails, cleaning up the database, generating pdfs) it's better to use resque or sidekiq
Now, I wonder if it is possible to do typical backend stuff like
sending (bulk) emails, cleaning up the database, generating pdfs, etc.
with those new frameworks. And in a way that is productive and doesn't
suck. I also wonder how difficult it is to create complex forms with
them. I got used to the convenient Rails view helpers and Ruby gems to
handle those kind of things.
Also, my question is not only about background jobs, but also about stuff one can might do during a request, like generating a pdf, or simply rendering complex views with rails helpers or code from gems. –
You're mixing metaphors here - a single page app is just a site where the content is loaded without doing a full page reload, be that a front end in pure js or you could use normal html and pjax.
The kind of things you are describing would be done in a background task regardless of the fornt-end framework you used. But +1 for sidekiq if you're using ruby.
As for notifying all the other users of things that have changed, you can look into using http://pusher.com or http://pubnub.com if you don't want to maintain a websocket server.

Rails Authentication via Web Service

So, this may be a kind of dumb question, but I checked the Google and got no hits. We want to host multiple Rails apps in a way that makes them look homogeneous. We want all the apps to have the same look and feel, and all the apps to use the same sign-on database.
Theming I think we could accomplish by just putting the site theme into a gem, and requiring that gem from our github repository in each app. However, auth is trickier.
I know that I can achieve this "for free" by just not making the different portions of the site (store, chat forums, etc.) different apps. If they're all, say, Rails Engines, we can basically drop them into the same application with their own namespaced routes, and have a single plugin that does auth.
However, for various reasons we'd like to keep these separate apps, if that's technically possible. The number one reason is scalability; since this will be a hosted site, we want the flexibility to spin up more instances of, say, the store (perhaps to handle a holiday sale rush), without needing to spin up the chat forums. Also, we want to be able to completely isolate the portions of the code that AREN'T intertwined.
Ideally, the databases would be separate too (keeping us from falling back into the rut of "put everything including the kitchen sink in the db"), but I do know that one "cheap" way to do cross-app auth is just to use the same plugin (say, Devise), and just point to the same DB.
So, I'm thinking that maybe the way to do this is to auth via a web service call. Is this prior art -- does anyone have a gem for this that "just works" so that authentication can be shared across all apps? Or am I just entering into a world of pain by trying to build things this way?
Thanks in advance!
You could do a single sign on approach described at:
http://blog.joshsoftware.com/2010/12/16/multiple-applications-with-devise-omniauth-and-single-sign-on/
The single sign on approach with oauth and devise has some drawbacks. The main problem I had was I was unable to extend the timeout time across multiple apps.

Scalability of multi-site rails app

I am beginning work on a new Rails project that is based on the premise of allowing users to create their own "sites." Each "site" would be a subdomain of the root domain (we'll use example.com). So if user Foo wants to create his own site at bar.example.com, each page request to a bar.example.com page would require fetching the a row in a sites table based on the subdomain.
My question is not how to code a multisite app, I think I have a pretty good grasp on that. My question is, from a scalability and performance perspective, would it be better to simply generate a new rails project for each site a user creates? Or is it ok run all sites out of one rails app. If numbers are necessary, let's assume I have 1 million users, each with a maximum of 5 sites, with each site bringing in around 1,000 hits a day.
I realize this is kind of a broad question, and mostly depends on my implementation of either method to reach a feasible solution, but any suggestions in terms of the best way to write this, including optimizing the DB, etc. would be appreciated.
It would be exponentially easier to have 1 rails app with millions of subdomains compared to millions of rails apps.
Check out this railscast for how to start with subdomains: http://railscasts.com/episodes/221-subdomains-in-rails-3
I wouldn't ever consider doing something like this with multiple Rails projects, because of the need to maintain all the code. By keeping it centralized, you can change the functionality of everybody's sites at once.
I think you might also run into memory issues by having all of those copies of Rails instantiated, too.
#Solomon is right. Heroku.com is using same concept for it's users to demonstrate users' applications.

How to extend an existing Ruby on Rails CMS to host multiple sites?

I am trying to build a CMS I can use to host multiple sites. I know I'm going to end up reinventing the wheel a million times with this project, so I'm thinking about extending an existing open source Ruby on Rails CMS to meet my needs.
One of those needs is to be able to run multiple sites, while using only one code-base. That way, when there's an update I want to make, I can update it in one place, and the change is reflected on all of the sites. I think that this will be able to scale by running multiple instances of the application.
I think that I can use the domain/subdomain to determine which data to display. For example, someone goes to subdomain1.mysite.com and the application looks in the database for the content for subdomain1.
The problem I see is with most pre-built CMS solutions, they are only designed to host one site, including the one I want to use. So the database is structured to work with one site. However, I had the idea that I could overcome this by "creating a new database" for each site, then specifying which database to connect to based on the domain/subdomain as I mentioned above.
I'm thinking of hosting this on Heroku, so I'm wondering what my options for this might be. I'm not very familiar with Amazon S3, or Amazon SimpleDB, but I feel like there's some sort of "cloud database" that would make this solution a lot more realistic, than creating a new MySQL database for each site.
What do you think? Am I thinking about this the wrong way? What advice do you have to offer in this area?
I've worked on a Rails app like this, and the way it was done there was named-based virtual hosts, with db entries for each site running. Each record was scoped to a site if necessary (blog posts, etc.) while users would have access to all sites running out of that db. Administrator permissions could be global or scoped to one or more sites.
You're absolutely correct when you say you'll reinvent the wheel a million times during the project. Plugins will likely require hacking on top of the CMS itself.
In my situation, it ended up being a waste of almost a million dollars of company money to build that codebase to run multiple sites while still being able to cater to the whims of each client site. It worked, but was not very maintainable due to the number of site-specific hacks that subsequently entered the codebase. You may be able to make it work if you don't have to worry about catering to specific client sites running on your platform.
In the end, you're going to need a layer of indirection to handle the different sites regardless of methodology. We ended up putting it in the database itself. If you go with the different-db-for-each-site method you mentioned, you'll put that layer in your code instead. I'm not sure which one is the better method.
I hope you're able to pull this off. I failed.
Also, as I learned today, Heroku offers postgres instead of mysql for rails apps.
There's James Stewart's Theme Support Plugin for Rails 2.3, and lucasefe's themes_for_rails gem for Rails 3+.
I just started using the 2.3 version and it's working well so far.

What is the best strategy to combine IntrAnet and Web-exposed website?

I was wondering if somebody has some insight on this issue.
A little background:
We've been using Rails to migrate from an old dBase and Visual Basic based system
to build internal company IntrAnet that does things like label printing,
invetory control, shipping, etc - basically an ERP
The Dilemma
Right now we need to replace an old customer-facing website that was done in Java, that
would connect to our internal system for our clients to use. We want to be able to pull information like inventory, order placement, account statements from our internal system and expose it to site live. The reason is that we take orders on the website, through fax & phone and sometimes we have walk-ins. So sometimes (very rarely thou) even a short delay in inventory update on our old Java site causes us to put an order on backorder, because we sell the same item to 2 customers within half an hour. It's usually fixed within one day but we want to avoid this in the future.
Actual Question
Does anyone have any suggestion on how to accomplish this in a better
way?
Here are three options that I see:
a) Build a separate Rails app on a web server, that will connect to the same DB that our internal app connects to.
+++ Pluses:Live data - same thing that our internal apps see, i.e. orders are created in real time, inventory is depleted right away
--- Minuses: Potential security risk, duplication of code - i.e. I need to duplicate all the controllers, models, views, etc. that deal with orders.
b) Build a separate Rails app on a web server, that will connect to a different DB from our internal app.
+++ Pluses: Less security exposure.
--- Minuses:Extra effort to sync web DB and internal DB (or using a web service like REST-API), extra code to handle inventory depletion and order # creation, duplication of code - i.e. I need to duplicate all the controllers, models, views, etc. that deal with orders.
c) Expose internal app to the web
+++ Pluses: all the problems from above eliminated. This is much "DRY"er method.
--- Minuses: A lot more security headaches. More complicated login systems - one for web & one for internal users using LDAP.
So any thoughts? Anyone had similar problem to solve? Please keep in mind that our company has limited resources - namely one developer that is dedicated to this. So this has to be one of those "right" and "smart" solutions, not "throw money/people/resources at this" solutions.
Thank you.
I would probably create separate controllers for the public site and use ActiveResource to pull data from you internal application. Take a look at
http://blog.rubybestpractices.com/posts/gregory/rails_modularity_1.html
http://api.rubyonrails.org/classes/ActiveResource/Base.html
Edit - fixed link and added api link
I would go for a. You should be able to create the controllers so that they are re-usable.
Internal users are as likely to duplicate data as external users.
It's likely that a public UI and an internal, for-the-staff, UI will need to be different. The data needs to be consistent so I would put quite a bit of effort into ensuring that there is exactly one, definitive database. So: one database two UIs?
Have a "service" layer that both UIs can use. If this was Java I would be pretty confident of getting the services done quickly. I wonder how easy it is in Ruby/Rails.
The best outcome would be that your existing Customer Java UI can be adapted to use the Rails service layer.
Assuming you trust your programmers to not accidentally expose things in the wrong place, the 'right' solution seems to me to have a single application, but two different sets of controllers and views, one for internal use, and one for public-facing. This will give you djna's idea of one database, two UIs.
As you say having two separate databases is going to involve a lot of duplication, as well as the problem of replication.
It doesn't make sense to me to have two totally separate apps using the same database; the ActiveRecord part of a Rails app is an abstraction of the database in Ruby code, therefore having two abstractions for a single database seems a bit wrong.
You can also then have common business rules in your models, to avoid code duplication across the two versions of the site.
If you don't completely trust your programmers, then Mike's ActiveResource approach is pretty good - it would make it a lot harder to expose things by accident (although ActiveResource is a lot less flexible and feature rich than ActiveRecord)
What version of Rails are you using? Since version 2.3 Rails Engines is included, this allows to share common code (models/views/controllers) in a Rails plugin.
See the Railscast for a short introduction.
I use it too. I have developed three applications for different clients, but with all the shared code in a plugin.

Resources