I started working on this site nearly four years ago. The idea was that it would be a blog about my bicycle trips. I eventually decided to just put everything here, moving away from my old site, mjhoy.com.
Finally, just a few days ago, I implemented automated deployment, which I’m mostly happy about, and thought I’d write about how this all works.
The code and the content for the site lives at github.com/mjhoy/mikey.bike. There is no database or runtime: I am ultimately just rsync’ing HTML and images and CSS up to my web server. These files are generated from a source of markdown files and templates by a Haskell project using the excellent hakyll library for generating static sites.
Hakyll lets me build up what is essentially a compiler for my site. I can define rules for taking input files and producing output files. In the simple case, I simply copy files. This rule says copy everything under
"images/**/*" $ do match route idRoute compile copyFileCompiler
route is a function for producing a path given the input path;
compile is a function for producing a file given the input file. I’m essentially using the identity function for these. (A more complicated version of this rule might pipe all my images through imagemagick to size down for the web.)
Hakyll also gives you tools to do more complicated things with, say, a series of journal entries, for you which you need an index and an RSS feed and a “landing” page and all that. For example, here is the code to produce my RSS feed. My journal entries are written in Markdown, with some metadata added to the top of the file (like the title and date).
When I push up a new commit, a Github action fires off to test the commit. It builds the Haskell project, and then attempts to build the site. Any Haskell compilation errors are caught by the first step, and any content issues are caught by the second – say I misspelled the “date” metadata field in a journal entry. The Github action for this is pretty simple. One important step is to set up caching of builds, because building all of Hakyll’s dependencies takes a long time – around 40 minutes! With the cache, if dependencies haven’t changed, each run takes only about 2 minutes.
Finally, I also wrote an action to deploy the site when a PR is merged into the master branch. It adds one additional step to deploy the site; after building, it just rsyncs the output to my web server. For this to work, I added several secret keys to a “deploy” environment I gave this action access to.
Overall, I’m pretty happy with this. Although I still edit my journal entries locally, I could also now use Github’s editor to add a new markdown journal entry, commit the change, and deploy the updated site to my server.