Fantastic episode, thank you.
I have an enterprise type question: If we take 37signals' Basecamp as an example of what we want to achieve in terms of separating customers logins and data by subdomain, presumably each database table would need a subdomain ID reference or else the whole thing would be a mess.
Is it normal to have one massive database (with the help of clustering, load balancing etc) or 1 database per subdomain (which presumably would be routing hell).
I just wondered what conventional wisdom is and how well this model scales?
@Mike, I believe Basecamp does not make separate tables or databases for each subdomain. It's just one database with multiple tables split by models just like Rails does.
Usually when you're working with subdomains like this, you'll have a model which maps directly to the subdomain which you load in a before filter like I do here. In Basecamp the model could be Company, not Blog.
You then always fetch other models through the associations as I show in the episode. This way they're always scoped properly and you can't access models outside of the company/blog.
Thanks for the awesome cast, I will be using this in an upcoming project :)
Great work ryan. I have everything working really well. Except for one thing. The conditions regular expression does not seem to be working. It still goes to the wrong url when there is no subdomain. I thought maybe it wasn't working so I tried it out. I hard coded a subdomain :subdomain=>/test/ and it worked as expected. I'm just curious if I am missing something. again thanks for all of your awesome free screencasts
subdomain-fu is pretty cool. We used it in http://hoptoadapp.com
One problem we had with subdomain-fu was that it broke autotest for the app. So, I forked and ham-handedly removed the autotest directory - http://is.gd/1IIT
On the topic of subdomains, I wrote (but haven't used yet in an app) a copycat of the excellent validates_email_format_of plugin for subdomains that someone might find useful - http://is.gd/10hL
I can't see the videocast in Linux (mplayer, vlc or totem) ... In fact I can see the video, but it's very slow, like 1 frame per second....
I also have problems viewing the movies in Linux. I have tried Kaffiene, VLC, xine, and mplayer. The problems started around episode 100. One player shows the video but doesn't play the sound, one plays the sound but the video is extremely choppy, one crashes and one does nothing. Maybe you changed some settings that affect the mov file?
I used to watch your screencasts religiously, but now I study the notes deeply instead. So your work is still incredibly useful to me, and engages my brain even more. I just don't get the full effect most of the time...
@jeff, what is the "wrong url"? It should be going to whatever you specify with "map.root" when not passing a subdomain.
@Nelson and Kurt, I'm hoping to release the episodes in a more Linux friendly format in the future. Thanks for your feedback.
Awesome - I've been looking forward to this one :) Very handy for lots of my future projects.
I've had no problems with railscasts on Linux. VLC doesn't work for me (I get audio, but the image freezes on the intro), but MPLayer works perfectly. (dev-SVN-r26753-4.1.2)
Thanks for the cast, Ryan.
Please don't use wildcard CNAMEs. Use wildcard A records, if you set up your own DNS server.
*.example.com. A 127.0.0.1
@Michael, thanks for catching that. Fixed the show notes.
I have the same issue as Jeff..(i think) No matter what I specify in the conditional regex, it matches and sends me to the show action instead of falling through to the map.root ..
example, both test1.project.local, test2.project.local and project.local takes me to the show action of the controller..
@kenny, hmm, that's strange. Try downloading the project for this episode and see if you get the same behavior there.
http://github.com/ryanb/railscasts-episodes/tree/master/episode-123
I actually have a question which I've been meaning to ask for a while. Why do you use build (as opposed to new or create). I know build just makes the object in memory.. but doesn't new do the same thing (until you save it)?
@Ryan:
I just tested your project from github, and it does work.. Very strange, I cannot see any differences compared to mine (the only difference is that I'm using my work computer now, and yesterday personal, comp. But both running leopard with passenger on original apache and so on). I will try again later today, doing once again from start and see what happens.
The latest commit of PassengerPane to GitHub includes an "Aliases" field that takes care of the ServerAlias addition to the vhost AND updates your hosts file.
@Marc, interesting, I didn't realize "new" worked when called through associations. But looks like it does from my brief testing. Anyone know if there are differences or side effects?
Still, "build" seems to be the convention so I'll probably stick with it, and "new" is kind of reserved for Ruby which is why I'm surprised it works.
I've actually taken this a step further by also rendering a different set of templates per blog.
So instead of rendering from /app/views, the app renders /app/views/domain. So it's one rails app serving n-blogs on full domains.
Get in tough with me if you want to share your thoughts on this topic.
@Michael Graff - I think that wildcard for subdomains is the best solution. For apps like basecamp, we have to authorize usera for some domain just at the begining when he is loggin and save subdomain it session. This better solution, dont you think?
I'm having the same problem as jeff and kenny above. Conditions on a route don't seem to be working. For instance, the following route:
map.account_root '',
:controller => 'account_root',
:action => 'index',
:conditions => { :subdomain => /.+/ }
keeps on redirecting to itself. The account_root controller tries to redirect to the index page if the current account cannot be found. With the conditions not working, it falls into a redirect loop.
Any ideas?
Ray, how would I accomplish the following:
I have a website 'xyz.com' that dynamically generates sites for social networks and assigns them each a unique subdomain, ie sn1.xyz.com, sn2.xyz.com etc.
The 'site' controller handles all 'xyz.com' requests without the subdomain whereas the 'network' controller handles all requests that have the subdomain present.
How can I conditionally route incoming request based on whether the subdomain is present or not?
We have a question on combining subdomains (done with SubdomainFu) and selenium-on-rails. Whenever we access a subdomain we get a same origin policy permissions error. How do we test an application using selenium that uses subdomains? Thanks!
@Marc, I tried MPlayer from the command line and it works fine. I guess I need to figure out what's different between clicking on an icon vs. running the program from the command line. Thanks for the clue.
I was having issues when trying to use the route in the screencast:
:conditions => { :subdomain => /.+/ }
This was not working for me when trying to hit the site using www.domain.com. "www" is ignored by default in subdomain_fu, but this condition seemed to treat the "www" like it was a true subdomain. The route I'm now using has this condition:
:conditions => { :subdomain => true }
This seems to work as expected with "www".
Very interesting screencast.
However after viewing it I have more questions ;)
You set up the blog as the subdomain and then in your controller you call something like @blog.posts.find(:id) in order to have only the posts associated with the blog. This is fine when you have few models and not really deep. But how would you do this if you have say comments on your articles? Checking if one comment belong to the blog is a bit more complicated. And if you can comment on the other's people comment, then it become really messy.
How would you do such a thing? I can't find a satisfying solution.
@jeff, @kenny, @erik
I had similar problems that you describe in your posts.
I found that I makes a difference whether you install it as a plugin (from github) or a gem. With the gem I could not get the route to accept the :subdomain-condition correctly. But then I installed the plugin with
./script/plugin install...
and all of a sudden it worked fine.
I'm workin on a debian-linux-machine (4.0) with Rails 2.1.0 if that is any help. I have no idea how this works on other setups. ;)
PS: don't forget the initializer-script for local development mode to set the number of tlds right.
@ryan: very nice screencast - I love your work. Thank you.
@keyJ,
I have tried both the gem and the plugin install, as I suspected that the problem was as you describe.. Still can't get it to work though..
@Sebastian
I never said wildcards were not a good use here (although I personally hate them) but a wildcard A record is safer to use, and more efficient than, a wildcard CNAME. For one, it is only one DNS lookup, where the other can be up to two (if the target of the CNAME is not the same as the parent. e.g., *.example.com CNAME www.example.ORG)
Wildcards in DNS don't often do what people expect. They don't behave like 'regular' wildcards, but as a strange beast at times. As an author of a DNS server (BIND 9) we often get bug reports about them. Use them carefully, is all I say!
That said, this use of them is common, and pretty tame.
We have used this in our latest project - but was wondering what SNAFUs, if any throwing in SSL into the mix - anyone has encountered?
Do I need to purchase a wildcard cert?
Anyone else experiencing problems when using this in combination with will_paginate? When I visit the site with subdomain.domain.local, all the url_for and link_to's are correctly pointing to subdomain.domain.local/path, except for the will_paginate links. will_paginate generates links to domain.local/path.
You can give railsolver a try for wildcarding hosts.
This plugin will "hijack" Ruby host(s) resolver aka(resolv-replace.rb), allowing programmatic host file resolution (including wildcard host resolution). ™
http://github.com/retr0h/railsolver/tree/master
Like several others in these comments, I had major problems with the routing working correctly like it does for Ryan in this screencast. FYI, I'm using webrick, and I'm on windows xp.
So I took Ryan's advice and installed his app.
What I discovered is that Ryan's app's routing worked at first, then stopped working once I added "www." in the main domain in my local machine's hosts file. The routing works fine for the blog app when you just have <code>127.0.0.1 personal.blog.local company.blog.local blog.local</code>. What's interesting is that I'm using webrick and not apache like Ryan is using in the screencast. You'll notice he didn't have to specify the main domain at all. If I leave out the main domain, I can't access it at all, so I have to specify it on my setup.
Taking out the "www." didn't fix my local app though, which drove me nuts for a long time. Then I tried installing the plugin instead of the gem I was using, and sure enough the routing started working properly. I don't know what the difference is, but I'm glad for routing that works now.
Summary: use the plugin, don't specify "www." for the main domain in your hosts file.
I had problems with the www working too. The plugin is awesome, but would be a lot better if this area could be cleaned up.
Initially I had to do this to get it to work:
# routes.rb
map.blog_root '', :controller => 'blogs', :action => 'show', :conditions => { :subdomain => /[^www]/i }
map.root :blogs
which regex matches when the subdomain is not "www"
Later I changed it to this which worked too and seemed simpler:
# routes.rb
map.root :blogs, :conditions => { :subdomain => false }
map.blog_root '', :controller => 'blogs', :action => 'show'
Notice the root has to come before blog_root in that scenario.
-Brian
StartBreakingFree.com
Thanks so much for this, Ryan! It came in really handy in one of my projects, cleaned it up right nice.
@Brian (#40):
Your first example wasn't matching what you think it was. The /[^www]/ regex is equivalent to /[^w]/, and matches any string containing a character that isn't a "w". So "www" would fail, but so would "w" or "ww" or "wwwww". Obviously a little contrived, but still not really what was intended. This seems to work, although it looks awfully hacky:
/^(?!www).+|www.+$/
Your second solution is much better.
Does anyone have a good resource for how to do domain mapping? Basically, the app knows accounts by subdomain, but the web server up the stack does the url rewriting. Haven't found a definitive bullet-proof resource on this.
Thanks,
Bill
@Loïc, good question. I would like to know what people think the best solution for this is too. Ryan answers Mike (comments 2 & 3) by suggesting 'always fetch other models through the associations'. Surely not always?! Is the answer to include the company/domain id in strategic places on the DB? ie Denormalise a bit.
@Ryan. Thanks!
I want to create a sub domain functionality for my project same as http://hoptoadapp.com/account/new. For that i have read this article(http://railscasts.com/episodes/123-subdomains) and try to do that. I just download the full source code "episode123/blogs" and try to run it on my windows platform. But i get an error message, here it is: ActionController::InvalidAuthenticityToken in BlogsController#create
ActionController::InvalidAuthenticityToken
Is their any solution then please let me inform.
platform:
windows
rails version:2.1.0
add 127.0.0.1 personal.blog.localhost to /etc/hosts file
Thanks,
Soumya
Hi,
I didn´t quite undestand if I need to modify the vhost.conf if i am using script/server.
Also i didnt find the file on my local machine, I tried with one subdomain and it worked, but when i write request.subdomains.inspect i get an empty array.
thanks.
Thanks for the great rails cast.
I am trying to install the plugin and it keeps coming up with hanging during the installation. Here is the output:
removing: /Users/Tim/Code/Ruby/RainbowCube/vendor/plugins/subdomain-fu/.git
Initialized empty Git repository in /Users/Tim/Code/Ruby/RainbowCube/vendor/plugins/subdomain-fu/.git/
fatal: exec index-pack failed.
fatal: index-pack failed
~/Code/Ruby/RainbowCube$ remote: Counting objects: 26, done.
remote: Compressing objects: 100% (19/19), done.
Can anyone help me??
Thanks,
Tim
Awesome. Thanks so much, Ryan. It seems as though whenever I need to do something which is new for me you have a screencast about it.
Keep up the great work. You rock! :-)
I just want to confirm that I was also having problems where, after setting up subdomain_fu the app seemed to think that 'www' was an account name and so through an error. I followed the advice of @bryce (#30 above) and simply changed:
:conditions => { :subdomain => /.+/ }
to
:conditions => { :subdomain => true }
This seems to be working fine - 'www.myapp.com' now defaults to the same place as 'myapp.com'
Thank's a lot Ryan for covering this.
I've been waiting for a time when I'll use subdomain. And this happened today.
Without this episode ...it's sad to imagine.
Thanks!
thanx for a nice article on the isssue of subdomains..Everyone has a "www" subdomain (In "www.yourdomain.com", the "www" part is the subdomain). The www subdomain is created by, and serves files from, your /htdocs/www directory on the filesystem.
To add more subdomains, just create new subdirectories in the /htdocs directory.
For instance, if you had the domain name mychevycars.com and wanted a subdomain for Corvettes, then you would just create the directory /htdocs/corvettes and upload the files in there that you wanted to see at http://corvettes.mychevycars.com
How can I make rails generate ALL urls to point to a valid subdomain?
Example:
personal.blog.local has a page with a url:
<%= link_to "Show blog", blog_root %>
If I follow your howto, this will work, obviously, but now my question is: how can I generate urls to, let's say the "edit" action?
<%= link_to "Show blog", edit_blog_root %> doesn't work. Kind of logical, since we didn't define it anywhere, but I still want to know how this can be done dynamically.
So the thing I'm looking for, is some kind of directive in the routes config file, that generates all kinds of default paths I can use. Something like this:
map.resources :blogs, [insert specific information here so that rails generates edit/destroy/show/etc... URLs]
Is this possible anyway?
Ryan, excellent screencast - thanks!
Like other commenters, I ran into an infinite redirect when when using the gem version and a non-valid subdomain on ubuntu and 2.2.2 and script/server.
The fix was simple. Use :subdomain => false on the redirect:
redirect_to root_url(:subdomain => false)
BTW as of today with rails 2.3.2 the session domain thing has apparently changed.
Now it should be ActionController::Base.session_options[:domain] = '.blog.local'
and not [:session_domain]
Thx for the cast!
Matthew Hollingworth has a great gem that plays well with routes. It's worth looking into - it avoids before filters or passing around :subdomain in url parameters.
http://code.matthewhollingworth.net/articles/2009-06-02-adding-subdomains-to-rails-routing
If you're running Rails 2.3.x and the latest version of subdomain-fu, make sure you set
SubdomainFu.override_only_path = true
in your initializer or else passing :subdomain into URLs won't work.
on iTunes
