Switching to Docker and CoreOS
I learned about Docker over the summer at ApacheCon in Denver. While Docker, itself, wasn’t on the program, it came up several times as various people were talking about PaaS systems. Once I started to dig into it, I understood why people were so excited. After playing with it more on my own, I was hooked. I decided that I wanted to move this site to Docker.
In this post I’ll tell you a bit about what I did, how I did it and why. What I’m not going to do is explain the full workings of Docker. If you want that, check out the Docker documentation.
What is Docker?
Solomon Hykes, on the Docker site, describes Docker thusly, “Docker is an open platform for developers and sysadmins to build, ship, and run distributed applications.” In other words, it’s a really convenient way to wrap up an application and everything it needs in one nice neat little package and run everything from within that package.
One of the things I like about it is that individual pieces can be isolated. For example, I write a lot of Perl and, as much as I like it, doing a lot with CPAN on a server can make a real mess. That’s especially true if you have multiple Perl apps using lots of different libraries. Keep all of that up-to-date and making sure a needed upgrade for one app doesn’t break another is time consuming and, frankly, not very fun. Docker allows you to keep each piece as separate form each other one as you want.
On top of all of that goodness, Dockers can be reliably replicated. I know that if I build a container and fully test it out on my laptop, that it work exactly the same when it’s deployed. That consistency is great when it comes to the whole DevOps thing.
Converting to Docker
Originally, this site was running on Nginx on an Ubuntu server in my living room. The web server worked fine (it’s a simple site after all) but the apps would sometimes freak out.
I use Octopress to generate the site and I have a couple of Perl scripts that do other things for me. That worked alright but Octopress is written is Ruby and the gem system is even more fussy than the CPAN. I don’t know how many times updates broke because something changed with a Ruby gem. Even worse was when I checked out the site from git on another box I have some sort of problem get the deploy step to work.
I had a great opportunity because I needed to move my site off of my home server. I decided to set up shop at Digital Ocean. DO is a great place to run your own virtual private server and they make it very easy to run certain applications like Wordpress and, more importantly for me, Docker out of the box. Their Docker application installs Docker on Ubuntu and is all ready to host Docker containers. Docker installs easily on Ubuntu even without their app but, hey, I’m all for making things easier on myself.
The first thing I needed to do was break down my site into pieces to Dockerize. The current Docker best practice is to have a each container do a single task. It this case, it was pretty simple to pick out those tasks. I would need four containers. The first container I’d need is nginx. Number two was for Octopress and three and four were for my Perl scripts.
I took a little bit of working with Docker files to make sure all of the needed libraries were installed for each app but that wasn’t hard. The real head scratcher was Octopress. You see, Octopress is designed to deploy the generated pages via rsync to a remote server. The rsync part is fine but I wasn’t going to running ssh or an rsync server in the nginx container just to publish updates. I had to hack on Octopress a little to allow to publish to a local directory and I was golden.
Now, let’s dig into the containers.
This is the easiest of the bunch. On my first pass, I created a Dockerfile which used the official nginx repository from the Docker Hub Registry. The only thing it changed was the location of the document root to match what was on my server. It turns out that that was a bad idea. It was easier to use the nginx repo unchanged and change apps to look to the new document root. It’s one less container for me to maintain and, thanks to other magic I’ll get to when I talk about CoreOS, it’s automatically updated.
I run the container with the following command:
--rm --name perlstalker_web-server -v /var/www:/usr/share/nginx/html
-p 80:80 nginx. There’s one piece of special magic in that I map
/var/www to the default nginx doc root
keeps the site data persistent even though the container is deleted
after every run and provides a nice hook for the other containers.
One of the first things I did to prep for this move (after I fixed my rsync issue) was to move my repo up to github. Now I had an easy way to get my site onto the server. The next step was to build the container.
Below is the Dockerfile.
FROM ubuntu:trusty MAINTAINER Randall Smith <[email protected]> RUN apt-get update RUN apt-get install -y git ruby ruby-dev gems rbenv ruby-redcloth build-essential python-pygments nodejs WORKDIR /usr/local/src RUN git clone https://github.com/PerlStalker/perlstalker.vuser.org.git perlstalker.vuser.org WORKDIR perlstalker.vuser.org RUN gem install bundler RUN rbenv rehash RUN bundle install ENTRYPOINT git pull && rake generate && rake deploy
When you run
docker build against this Dockerfile, it will install
all of the necessary requirements, clones the site from github and
then finishes the install. Once that’s complete, running the container
will pull the latest updates from github, generate the static pages
and deploy the site into the doc root.
The cool thing is that this container can be built once and run as
often as required. (I know. I’m easily amused.) The running containers
can even be removed on completion (with the
--rm flag to
run) and re-run.
The other trick is to mount the document root from the nginx container
so that the generated files from Octopress get put in the right
place. Use the
--volumes-from flag like so:
--rm --name perlstalker_deploy --volumes-from perlstalker_web-server
One of the scripts I use on my site generates a RDF feed for my daily scripture study. The code, including the Dockerfile, is up on github. I’m not going to go into details. You can check out the Dockerfile for yourself. Again, the trick is that it mounts the doc root from nginx container.
I ran on Ubuntu for a while but ran into an annoying problem when it came to updates. You seen, Digital Ocean used an external kernel when starting Ubuntu VMs. It’s great, in one way, because it’s really fast to start up. On the other hand, it causes no end of problems when working with Docker. I frequently ran into issues where I would forget to change the kernel I was booting with to match what was just installed by apt. Sometimes the VM wouldn’t boot, other times docker refused to start.
The other catch is that, to be honest, I got really tired of applying patches to an OS that isn’t really doing anything. All of the fun stuff happens in Docker. I didn’t exactly want to have to worry about Ubuntu.
Fortunately, Digital Ocean rolled out the option to create CoreOS VMs. I decided to get in on that action despite the beta status.
The big problem converting to CoreOS is that I had to learn systemd. That was annoying but wasn’t too bad. I’d like to share a couple of the units to show the magic.
The first is the main web server.
[Unit] Description=perlstalker.vuser.org web server After=docker.service Requires=docker.service [Service] TimeoutStartSec=0 ExecStartPre=-/usr/bin/mkdir /var/www ExecStartPre=/usr/bin/docker pull nginx ExecStart=/usr/bin/docker run --rm --name perlstalker_web-server -v /var/www:/usr/share/nginx/html -p 80:80 nginx [Install] WantedBy=multi-user.target
I want to draw your attention to the
ExecStartPre lines. The first
one creates the persistent storage for the web site pages. The
prefix tells to systemd to ignore errors like, for example, the
directory already exists.
It’s the second one that’s interesting. Every time the service restarts, it pulls an updated nginx image. That means that every time my server reboots or the service is restarted, I get a fully updated, patched and, theoretically, secure web server.
The next big piece is the Octopress deployment.
[Unit] Description=Generate and deploy site After=perlstalker.service Requires=docker.service [Service] TimeoutStartSec=0 RemainAfterExit=no Type=simple ExecStartPre=/usr/bin/docker pull perlstalker/sysadmin-deploy ExecStart=/usr/bin/docker run --rm --name perlstalker_deploy --volumes-from perlstalker_web-server perlstalker/sysadmin-deploy
Now, every time I run
systemctl start deploy, the Octopress
Docker updates the site.
I want to show you two last systemd units which trigger the update of my scriptures feed, in part, because I want to remember to crazy way systemd replaces cron.
Every cron needs to two units. One is the
.service file which
defines what actually happens.
[Unit] Description=Generate scriptures feed After=perlstalker.service Requires=docker.service [Service] TimeoutStartSec=0 RemainAfterExit=no Type=simple ExecStartPre=/usr/bin/docker pull perlstalker/scripture-feed ExecStart=/usr/bin/docker run --rm --name perlstalker_scriptures --volumes-from perlstalker_web-server perlstalker/scripture-feed
The second one is a
.timer file which sets the schedule.
[Unit] Description=Generate scriptures feed [Timer] OnBootSec=25min OnCalendar=*-*-* 04:30:00 Persistent=true [Install] WantedBy=timers.target
Make sure that you run
systemd enable scriptures.timer and
start scriptures.timer. I forgot to do that then wondered why my feed
didn’t update. :-)
I want to make a point here that may have been lost in my digression into systemd mazes. I didn’t have to change any of containers. I simply plugged them into the systemd on CoreOS and my site was flying again. If at some point, I decide to move to some other service such as GCE or EC2, I can drop my containers in, easy peasy.
Anyway, the point of this little screed is to show how a few building blocks or, shall I say, containers, can be stacked together to build whatever you need. Even if it’s something as trivial as my little site.