My team and I ditched the jekyll-assets gem for a homegrown asset pipeline and decreased build times by a factor of five. This is how we did it.
The background to this story:
My team and I were asked to use Jekyll on a handful of new projects that would transition one of our clients from an Angular application to the Jamstack. One of these projects has hundreds of pages, while another has thousands. While we overcame several build performance hurdles during the process, perhaps none were more significant than the issue with faced in building our front-end assets. So let's talk about that, shall we?
This gem seemed to be far-and-away the most popular for working with assets inside Jekyll. And, as a bonus, it was built atop Sprockets, which my team of Ruby on Rails developers really liked. Thus, we were off to the races using jekyll-assets.
(I'd bet you can guess what happens next.)
As the project matured, build times slowed. The project with hundreds of pages and the other with thousands of pages were both taking about 30 seconds to build in development. This is significant because it meant every time a developer made a a change, they'd have to wait 30 seconds to see that change on screen. Even if it was just a simple text change.
While a big part of this was a direct result of Jekyll re-building the entire site on every change, we were sure there was something we were doing to slow it down further. Or, perhaps better stated, there was something we could do to decrease the build times.
As you may imagine, our developers were not happy (even though every one of us contributed to the lagging builds). We knew there were several factors as to why the builds were slowing, and we knew there were many approaches we could take to speed it up. Amidst conversations on those approaches, a theory emerged that jekyll-assets was a major problem.
So we did a test. We ripped out jekyll-assets along with all of our front-end assets, such that all we had left were a bunch of disgusting Liquid templates. We ran a build and ... it took about five seconds. If my elementary math skills are serving me right, that's about six times faster without and asset pipeline (i.e. without jekyll-assets). Or, said another way, jekyll-assets was taking about 85% of the total build time. On a site with thousands of pages, that's completely unacceptable.
The next step seemed like it should be to look for another asset pipeline gem. The problem is all the gems we could find were missing a component we believed critical to the success of an asset pipeline: It shouldn't run if it doesn't have to.
We developed a hypothesis: If we built an asset pipeline that ran conditionally based on developer behavior and the availability of previously built assets, we could decrease most local build times by a significant factor.
We also had a funny feeling that even when the asset build ran it wouldn't take 85% of the build time, but we couldn't know for sure until we built the thing.
So we set out to build a custom asset pipeline for these Jekyll projects. And since we were doing it anyways, we worked a couple extra goals into the project, ultimately ending up with this list:
<link>tags in liquid templates.
We ended up creating a new Ruby gem with these three features:
package.jsonscript. But it needs some help to hook into the Jekyll build process.
<link>tags for our liquid templates. This must receive the cache digest from the hooks to ensure the filenames are correct.
We proved our hypothesis true! When the asset build didn't need to run because the files were already in place, the build took about six seconds—about a second for the tags and hooks to run—speeding up most builds by a factor of five.
To our surprise, the build process with Gulp was quite efficient and, when the build did run, it would increase the total Jekyll build time to about 10 seconds, still faster than our original build by a factor of three.
This gem is an open source gem and is available at https://github.com/crdschurch/jekyll-asset-pipeline for use in any project. It unfortunately shares the name of another gem (oops!) so it must be installed by pointing to the GitHub repository, but it's there for the taking. (Further documentation for the gem is in its README.)
It is opinionated to serve the two projects mentioned here, and it's a little rough around the edges. But I imagine you will find some benefit to it. And if you feel like making a change to it, well, that's the value of open source projects.