Automagic Multisite Server with Jekyll and Nginx on a DigitalOcean Droplet
I have a bunch of small sites that are well suited to using Jekyll, but I didn’t have a bonehead easy way to get them online. My non-solution so far has been to have a DreamHost VPS, and just spin up a WordPress site on that badboy whenever I want a throwaway site. But that’s expensive, inflexible, totally uncool, and it’s not Jekyll so it’s slow, especially if you get hit with a lot of traffic at once.
But it sure is easy.
So what I wanted is an expandable server that:
- Can have an arbitrary number of sites on it, that
- will automatically update any of the sites I assign to it,
- without me doing anything beyond making the changes and pushing to github.
- Bonus points if it can set up a new site, as well as updating existing ones.
So what I wanted to do is:
- Spin up a tiny Droplet on DigitalOcean (that’s about $5 a month)
- Set up nginx on it to serve more than one site
- Create a shell script that generates a new server block (aka virtual host) given just the new domain
- Write a tiny Sinatra app to act as a github webhook endpoint. The endpoint should:
- Clone the updated site to a temp directory
- Create a new server block in nginx if the site doesn’t already exist
- Create a new DNS record if the site doesn’t already exist
- Use Jekyll to build the files and put them in the right public_html directory
Note: If you have no idea what github, VPS, nginx, shell script, sinatra, etc mean, then this tutorial is not for you. If you’ve never used a linux command line, this tutorial is not for you. I try to be thorough here, but you’re going to get lost unless you have a grasp of the technology here.
What a shitshow this all was to get set up, but it fully works, so here’s the scoop:
This is the easy part.
Go into the DigitalOcean control panel and create a droplet. I used Ubuntu 13.04 x64. This should all work in theory in other versions, but if you want to be safe, use the same.
Note the IP it gives you because we’ll be using it (for this tutorial I’ll refer to that IP as XXX.XXX.XXX.XXX). Serving html files only is the easiest job a server can possibly do, so a small droplet will be fine unless you’re expecting massive traffic.
Note: You don’t have to follow this part strictly if you really know what you’re doing. If you’re not 100% sure you know what you’re doing and that you know enough to modify my instructions where appropriate, then just follow this section exactly.
For the purposes of this tutorial, I’m going to assume your domain is mysite.com.
You should own mysite.com through a service like NameCheap, and from NameCheap (or whatever service you use), you should assign the DigitalOcean name servers to the domain. (The name servers are ns1.digitalocean.com, ns2.digitalocean.com, ns3.digitalocean.com).
Then, in the DigitalOcean control panel, click DNS and add mysite.com, pointing at the droplet IP you just got.
This will give you a domain you can work with throughout this tutorial, and it can act as a “base domain” for later when we set up a git webhook.
Accessing the Droplet
Almost everything from now on will happen on the command line:
It’ll ask you if you trust the host, and you’ll say “yes”. Then it’ll ask you for your root password. DigitalOcean created that password when you created the droplet, and they sent it to you via e-mail, so check your inbox.
Once you’re in, first you probably want to change the root password to something secure, that you can remember:
Installing the Prerequisites
You’ll need some software installed on the server: nginx, ruby, rubygems, and a few gems including jekyll.
First, update apt-get
Then install Nginx:
That was really tough, I know, but stick with me.
We’re going to use RVM to install ruby:
Now, to actually use RVM we’ll have to use the source command. You might add this to your bash profile, but you’re not going to use rvm after this so it’s probably not worth adding. If you don’t know what any of that means, don’t sweat it.
Normally rvm sits at ~/.rvm/scripts/rvm so that’s the path I’ll use. If it gives you an error, then you need to figure out where your /.rvm directory is and change the following command to that path instead:
So now that you have access to the rvm command, you’re going to use it to install ruby and rubygems:
Now you have ruby and rubygems ready to go. You can check by typing “ruby -v” into the command line. We’re going to install a few gems we’ll use later:
If you want VIM or whatever, now’s the time to get it. I’m going to act like you’re using nano just because nano comes default, but feel free to use whatever.
Note: If you don’t know what VIM or nano are, you just need to know that they are text editors. Nano comes with Ubuntu, and it’ll work fine, but you can actually use whatever editor you want, if you have a favorite.
We’re going to manually create a site that nginx can serve, just to make sure it’s all working. First we’ll create the folder it’ll serve from, then we’ll set the right permissions:
Next you need to create an index file for nginx to serve out of your new directory:
Save that file and exit.
Set up the Test Server Block
Nginx comes with a default site configuration that you can copy. That’ll work for this test site, even though you don’t want to use it in the long term. We’ll use it for now and blow it away later:
Find the section of the file that looks like this, and make the appropriate changes:
Save and Exit.
You’ve just edited a file called /etc/nginx/sites-available/mysite.com. The sites-available directory holds “potential” sites, meaning sites that could be on the server, but that aren’t there yet. To make nginx actually serve the site you have to put the file into the sites-enabled directory. The best practice is actually not to copy the file, but to create a symlink there instead, to keep things DRY:
To avoid conflicts, you should also delete the default file you copied earlier:
Finally, restart nginx so the changes take effect:
Note: you should see a nice message from nginx about restarting. If you see nothing, then something is wrong. To check what’s wrong type tail /var/log/nginx/error.log. You’ll see the end of the error log and it’ll tell you what’s wrong. Probably you screwed up the server block config file.
In your browser, hit mysite.com. Now you have nginx serving a static page. Ah yiss.
The Server Block Script
I’m going to do things a little out of order here because some of the directories we need don’t exist yet, but we need the script that doesn’t exist yet to create them. So first:
In that file, put this code:
This script essentially does everything we did manually while we created the test mysite.com block. Nginx gets all the config files and directories set up correctly to begin serving the new site, plus a placeholder index file so you can make sure it works.
One other thing we need to do. Our test site works fine, but if we have more than one site on our server, it’ll fail to start because of a setting we need to change:
In this file you’ll find a line that has a setting called server_names_hash_bucket_size that is either commented out or set to 32. Change it to 64.
My explanation of what server_names_hash_bucket_size does isn’t quite right. It’s not that that setting allows for more than one site, but mysite.com is probably less than 32 characters long, whereas subdomains like thissubdomain.mysite.com can get go past the 32 character limit. server_names_hash_bucket_size must be set to a number larger than the largest domain name you have a block for.
So, let me get a little ahead of myself and ask you to run the new script for the git.mysite.com domain:
Nginx may complain that it can’t restart at this point, which is fine because the problem it’s complaining about is what we’re about to fix. We’re not actually ready to deal with the git subdomain yet, but having the folder there is useful, mostly because it gives us a place to put the script we just created:
Now that you have the script to create a server block, I want to give you a script that deletes existing ones, because:
- We’re going to kill the mysite.com test block because the server block configuration screws everything up if you have more than one site (that’s why nginx failed to restart just now when you created git.mysite.com).
- If your experience is anything like mine, you’ll be backtracking a whole lot, and this just makes it easier.
From the /var/www/git.mysite.com directory:
And use this code:
Save and Exit.
It just removes the server block configuration, deletes the www folder, then restarts nginx so the changes take effect. Run it:
That will delete the original test mysite.com block, and recreate it in a way that won’t bungle everything. mysite.com should be accessible from the browser right now. git.mysite.com is theoretically available, but we haven’t set up the DNS record for it to work yet. Let’s do that.
The hook is tricky because we’re on a server that only has nginx, which generally speaking can’t run scripts at all. We have to have a script though, in order to respond when github tells us that a site has been updated. Since we’re already using ruby for Jekyll, it makes sense to use Sinatra, a light weight ruby web framework.
We already set up our hook endpoint in nginx–that’s what git.mysite.com is for. We don’t have the DNS automatically updating yet, so let’s make that actually work by setting up the DNS ourselves.
In the DigitalOcean control panel, click DNS on the left. Click the View icon to the right of mysite.com, under Domains.
Click the large Add Record button near the top. Select CNAME as the type, then input “git” as the name and “@” as the hostname. That just tells the subdomain “git” to redirect to the same place as the A record for this domain, which is the droplet IP address.
Click Create and you’re done. It shouldn’t take long to propagate, so soon you should be able to visit git.mysite.com and see the test page we set up before.
Now, since git.mysite.com is going to serve our tiny Sinatra app instead of Jekyll sites, we need to edit the configuration manually:
Here’s the new configuration:
Save and Exit.
Basically we’re just passing any requests to git.mysite.com/ along to the port that Sinatra will be listening on (4567 is the default Sinatra port).
If that’s confusing, then think of it this way: nginx will receive a request at port 80 from the outside world. If that request meets certain criteria, nginx will make its own request to the local server, but on the port that Sinatra is listening on. Sinatra will process the request, and pass the result back to nginx, and nginx will pass the result back to whoever asked for it in the first place (probably github).
You can also delete /var/www/git.mysite.com/public_html if you want, since we only need /var/www/git.mysite.com. Don’t delete the log directory or nginx won’t be able to restart.
Now that we have nginx ready to pass a request on to Sinatra, let’s set up Sinatra.
Early in this tutorial you installed a bunch of gems: jekyll, sinatra, thin, json. This part is why you did all that.
Create a file in /var/www/git.mysite.com/ called githook.rb, with the following content:
Be sure to insert your actual github username in there. This script gets the latest site code, creates the nginx site if necessary, and builds from source using Jekyll.
Note: No github password is necessary here because cloning a public github repository doesn’t require a login. This works fine if all your sites are in public repositories. If you want to have a site in a private repository, then you’ll need to configure this server with your deployment user’s public key so you don’t have to type a password, otherwise this script will fail. If you don’t know what I’m talking about, just make your site’s github repository public. If you want it to be private, there are lots of tutorials about how to set up authentication with keys. Godspeed.
Note 2: One of my stated goals was to make the DNS update itself automatically if the record didn’t already exist. This script does not do that yet. I’m working on a ruby wrapper for the DigitalOcean API, but it’s not ready for prime time yet. So when you know you’re uploading a new site to the server you have to go into the DigitalOcean control panel and manually add the DNS information. You only need to do this once per site and only for new sites. When the script is ready, I’ll update this post.
If you hit git.mysite.com in your browser right now you’ll get a 502 Bad Gateway error (if you’re not seeing that, try restarting nginx). The reason is that right now nginx is trying to pass requests to http://localhost:4567 which is where Sinatra should be listening. But Sinatra isn’t running!
Run the sinatra server simply by typing:
You’ll see the session attach to a Thin server process. Success.
So we have a droplet that can listen for changes in a repository and update or create a site on the server in response.
The whole idea here is that we can use github to update our droplet, so we need to tell github how to contact us.
One important thing to note is that this whole process works by convention: your github repository must be named the exact site domain like mysite.com or sub.mysite.com – whatever you want to type into the browser to pull up the site is what you should name the github repository.
Now, create a new repository with the appropriate name. When it’s done, go to the repository’s settings, and click “Service Hooks” on the left. At the top of the long list you’ll see “WebHook URLs.” Click that. In the text field provided, type your git endpoint url: http://git.mysite.com.
Now, if I were clever I would tell you to clone the new repository on your local, and generate the jekyll file structure inside that directory. Jekyll doesn’t really like to do that though. It works like this:
Which creates a mysite.com directory with all the stuff in it. It’s possible to generate the files in a tmp directory, then copy them into an existing directory, but seriously man?
So here’s the workaround. Once you create the github repository, type this into your local:
Now your jekyll site is connected to the github repository, and guess what? When you pushed to origin, github told your webhook url about it, and your webhook url built the site on your server. It’s sitting there right now, waiting for you to load it up.
From now on, all you need to do to create a new website is:
- Set up the domain via the DigitalOcean control panel
- Create new github repo with name of site (e.g. petemichaud.com)
- Set post commit hook on repo to your http://git.mysite.com endpoint
- Generate a new Jekyll site locally
- Attach the new site to remote github repository
- Push generated files to origin
And from then on, every time you push new content to github, the site will be automatically in seconds. How cool is that?
I want to thank a few people for their help getting all this working: