Help me manage changes to a website

August 7, 2009 10:33 AM

Whats the best way to set up version control for a website?

I work for a small building supply company, and we have a custom online store which was developed by an outside consultant. He used to provide a 'development' site where we could make changes, which we would submit through subversion, and which would be reflected when we browsed to the development site.

He is no longer providing this service. I've set up a separate hosting account with dreamhost, since they provide subversion repositories as part of their package. I can upload our code to this hosting account and it works, and I can set up a subversion repository, but I can't get subversion on a 'live' site.

Can anyone provide pointers on how to get what I'm trying to do working, or suggest an alternative approach?

posted by Reverend John to computers & internet (6 comments total)

You can run svn out of cron:

* * * * * sh -c 'cd /var/www/your-site; svn update' 2>&1 >/tmp/svn-up-log

We do this for a few sites, and it works. But is a horrible idea. Developing locally and committing when your changes actually work is the best strategy. This is really easy to do in most cases; be it Perl, PHP, J2EE, whatever.

Also, svn likes to break in horribly strange ways. If you can manage, I would switch to git as soon as possible. It is much more reliable (and faster).
posted by jrockway at 10:38 AM on August 7

If you can manage, I would switch to git as soon as possible. It is much more reliable (and faster).

I would also suggest investigating git, but not for reliability reasons -- I use Subversion regularly in a production environment and I have never had any data loss or corruption since they moved away from BerkeleyDB storage many years ago.

The best reason for moving to git -- or any of a number of other version control systems with a distributed model (bzr, mercurial) -- is that they allow you to commit changes to a *local* repository, while only updating your production website when you're comfortable that all the changes are working and stable. So you get a complete revision history and the ability to revert changes at a very granular level while at the same time you can be sure that your website is always functional.

You may be able to achieve the same sort of thing with Subversion and branches, but it's by no means as simple or effective.
posted by larsks at 11:41 AM on August 7

I agree with jrockway, that Subversion can end up being a pain. I prefer Mercurial over Git but, of course, that's a personal preference thing. Just throwing out another option.

You can either use the copy/move/linked/updated method as described by him or find some kind of "continuous integration" method (Apache Continuum?) ... although that does seem overkill unless you're doing actual builds (e.g. with something like Maven)

I also agree developing locally is a little better for this kind of thing. It's either that or some kind of content management system that has plugins for an online store (and thus all the tedious work that goes along with it.)

Good luck with whatever you end up doing.
posted by jlstitt at 11:47 AM on August 7

I use subversion with a web-based application in a very similar scenario. There's the live site, there's a development version (which is often just a few steps ahead of the live site), and there's a local copy of the code that I work on.

What I've done for the development site is tie it to the working copy in SVN with a post-commit hook such that when I commit changes to my working copy, the development website shows the most recent changes.

So, for example the repository is laid out roughly as:
/trunk
/branches
/releases/v1.0
/releases/v2.0
/releases/...


I check out a copy of /trunk and make changes, then commit those changes back to /trunk. The post-commit hook takes the latest copy of /trunk and puts it out on the web-accessible development copy at, say http://development.example.com/. Since this is done automatically I can easily preview changes by committing code and then loading the development website.

When /trunk reaches a mature state I create an svn copy of it and place it in /releases, then I manually copy the current mature release into the production web-accessible copy at http://production.example.com/
posted by odinsdream at 12:04 PM on August 7 [1 favorite]

Here's the important part of my post-commit file:

/usr/bin/svn update /var/www/dev >> /var/svn/project/logs/post-commit.log
cd /var/www/dev && symfony cc

This goes in the hooks directory of your repository. Note that the first line will only work if /var/www/dev already contains a checked-out copy of /var/svn/project/trunk (or wherever your development code lives).

The second line clears out the cache created by symfony, the development framework used by my particular project. You could add additional commands here, like reloading apache config, say.
posted by odinsdream at 12:11 PM on August 7

I would recommend using something like Capistrano (or Vlad the Deployer). When you type a command such as 'cap deploy' they'll automatically check out the latest code from Subversion and deploy it to the live site. If anything goes wrong mid-deploy they'll put back the original site, or you can type 'cap rollback' and the previous site is restored. Capistrano is focused on deploying Ruby applications but can be used for plain HTML or PHP sites too.

They'll take a bit of time to set up but the end result is worth it.
posted by BinaryApe at 6:59 AM on August 9