Nextcloud's official Docker image is pretty large. With a little effort, can I reduce that by about 50%? Let's see!
Ever since its roots as ownCloud within the KDE project, I have always been fan of Nextcloud. On a personal level, I am also a fan of the team behind it. Back in my Kolab Systems days I even advocated an integration of Nextcloud (then ownCloud) and Kolab as strong, Free alternative to Sharepoint/Google Apps.
Since I am currently looking for new work I wanted to spend some time improving some skills. In this case, learning more about building Docker containers.
I picked Nextcloud to work with because it is non-trivial, but not a nightmare, to setup from source. There is also an existing official Docker image for Nextcloud which gave me something to work from.
Easy enough to get hold of:
docker pull nextcloud
Simply running this command immediately gave me an idea of where I might learn a ting or two: the image is over 600mb! That is massive. Unnecessarily massive.
So there was my challenge:
On first glance, there were a couple of things that immediately struck me as things that might be improved:
The Nextcloud image is based upon an official PHP image which, in turn, has Debian Jessie as a base image. I wanted to change this to Alpine Linux because it has a tiny base image (it is under 4mb!). This, however, forced me to look into how Alpine packaging works.
The Nextcloud image uses pecl to build/install some PHP extensions for Redis, memcached and apcu. For many reasons, I wanted to switch this to using the base distribution's own packaged version. Most importantly, I knew very little about pecl and how (if at all) it cleans up sources after building them.
My Dockerfile can be found on Github. At this stage it is still not complete enough to use (although it does provide a fully-working Nextcloud instance), so I am not going to make it available on Docker Hub. That said, even incomplete, there are some nice results:
My original mission was to see if I could reduce the enormous 600mb+ image size. Given my lack of experience, I would have been happy to shave off a few mb. I never expected this...
My image (at the top) represents a ~45% disk usage reduction over the official image!
Most of the size benefits came from using Alpine as my base image, I'm sure. I suspect there was some benefit in ditching pecl, too. Here be dragons...
In order to ditch pecl, I used Alpine native packages for the PHP extensions. Sadly, two of these packages (redis and memcached) are only available in the testing repository for Alpine Edge (not the stable distribution). So the benefits of native packaging are totally offset by the overall image potentially being less stable.
Alpine Edge also does not seem to have a php7-imagick package. So I had to borrow the php5 package. I assume this does not entirely work.
So this work is not entirely finished:
The stuff I can do will really not take very long; I should get that wrapped up in the next couple of days. However, we will need to wait for Alpine Edge to stabilise before I could actually recommend anyone using my Dockerfile in production.
Meanwhile, feel free to have a play with my Dockerfile and let me know what you think!