vampysite/source/_pages/JekyllOnNix.md

16 KiB

title date
Running a Jekyll website on Nix 2023-02-15 20:56 +0100

If you're reading this on technogothic.net and the year is something like 2023, this site is probably generated by Jekyll. Previously this website was hosted in an Arch Linux VM, then in a Debian 11 LXC container on top of Proxmox, but on both systems it was deployed and built manually with scp and bundler. The previous two systems became more and more unmaintainable over the years, so a few months ago I started migrating my servers to NixOS. This website was the last thing I migrated, and one of the most painful parts, by far.

Previous attempts

So, as with all things, I started by typing build jekyll website on nix into my browser's search bar and pressing Enter.
The first results that came up were:

To quickly summarize these articles, they have these steps in common:

First you have to run these commands:

bundler lock
bundler package --path vendor/cache --no-install
nix-shell -p bundix
bundix

This creates a bundler lockfile, fetches the dependencies into vendor/cache (without installing) and uses bundix to generate gemset.nix, a file used by Nix's bundlerEnv to know which gems we need. Then, we create default.nix, where we define the environment that nix-shell will later drop us into:

with (import <nixpkgs> {});
let
  env = bundlerEnv {
    name = "your-package";
    inherit ruby;
    gemfile = ./Gemfile;
    lockfile = ./Gemfile.lock;
    gemset = ./gemset.nix;
  };
in stdenv.mkDerivation {
  name = "your-package";
  buildInputs = [env ruby];
}

Clearly, these steps worked at some point between 2016 and 2022, as the blog post authors probably wouldn't have published these posts if it didn't. So, sounds simple enough, right?



Wrong

My first attempt

{ pkgs, lib, ... }:
let
  version = "a6476f6fc94f4476f31ebd35687dd54a807d41dd";
  repo = pkgs.fetchgit {
    url = "https://git.lain.faith/sorceress/vampysite.git";
    rev = version;
    sha256 = "1lr5jqsqik2czgas7ci4fynavhkcw6nfiw1p3lwlxm0ar3y483fk";
  };

  jekyll_env = patched_pkgs.bundlerEnv {
    name = "jekyll_env";
    ruby = pkgs.ruby;

    gemdir = "${repo}/.";
  };
in pkgs.stdenv.mkDerivation {
  inherit version;
  name = "vampysite";

  src = repo;

  buildInputs = with pkgs; [
    jekyll_env
    jekyll_env.wrappedRuby
  ];

  buildPhase = ''
    bundle exec jekyll build 
  '';

  installPhase = ''
    mkdir -p $out
    cp -r _site/* $out/
  '';
}

Time-traveling hashes aside, this was the first version that more or less looked like it could maybe work. The main differences here are:

  • I would like to package the website as a derivation to serve from my NixOS host, not have a development environment for Jekyll (though, after getting this to work, making one was quite easy as well).
  • The sources for the website are fetched from git, rather than being in the same directory.
  • It's now possible to specify just the gemdir in bundlerEnv without manually writing out each path.

The first issue I ran into was building the native extensions for the Nokogiri gem. It was missing a dependency, so let's add that to gemset.nix:

  mini_portile2 = {
    groups = [ "default" ];
    platforms = [ ];
    source = {
      remotes = [ "https://rubygems.org" ];
      sha256 =
        "b70e325e37a378aea68b6d78c9cdd060c66cbd2bef558d8f13a6af05b3f2c4a9";
      type = "gem";
    };
    version = "2.8.1";
  };

And then change the nokogiri definition above:

  nokogiri = {
-   dependencies = [ "racc" ];
+   dependencies = [ "racc" "mini_portile2" ];
    groups = [ "default" ];
    platforms = [ ];
    source = {
      remotes = [ "https://rubygems.org" ];
      sha256 = "sts693acKc131fOc09C2WrEJdb3s8EvnHWg/nJq+JmM=";
      type = "gem";
    };
    version = "1.14.1";
  };

I also added Nokogiri's dependencies to my build inputs:

  buildInputs = with pkgs; [
    jekyll_env
    # nokogiri dependencies
    zlib
    libiconv
    libxml2
    libxslt
  ];

Other than that, I had to retry building a few times and fix up some incorrect sha256 hashes outputted by bundix. Now that Nokogiri builds, the next issue I ran into was building sass-embedded:

building '/nix/store/04rabbrj3wgk7nxahz87wq0nsnyzl9c7-ruby2.7.6-sass-embedded-1.58.0.drv'...
  ...
Building native extensions. This could take a while...
ERROR:  Error installing /nix/store/x4dvi7mffypdbm3755vy221hjc2bapd9-sass-embedded-1.58.0.gem:
        ERROR: Failed to build gem native extension.

    current directory: /nix/store/rf9dnjclddg2ksyhycakq6jizkshv7dx-ruby2.7.6-sass-embedded-1.58.0/lib/ruby/gems/2.7.0/gems/sass-embedded-1.58.0/ext/sass
  ...
fetch https://github.com/sass/dart-sass-embedded/releases/download/1.58.0/sass_embedded-1.58.0-linux-x64.tar.gz
rake aborted!
SocketError: Failed to open TCP connection to github.com:443 (getaddrinfo: Temporary failure in name resolution)
  ...

Caused by:
SocketError: getaddrinfo: Temporary failure in name resolution
  ...

The build fails due to the gem Rakefile trying to download a tarball from GitHub during build time, which is not allowed in Nix derivations (and quite a lot of build systems, actually). As I found out by opening this issue, it's possible to override the path for the tarball and avoid the download. As this was ~6 hours into the process, I decided to not bother with it for now and instead went with a simpler workaround:

- gem "jekyll", "~> 4.3"
+ gem "jekyll", "~> 4.2.2"
  gem "jekyll-toc", "~> 0.17.1"
  gem "kramdown-math-katex", "~> 1.0"
- gem "rouge", "~> 4.0"
+ gem "rouge", "~> 3.0"
  gem "webrick", "~> 1.7"
  gem "image_optim", "~> 0.31.2"
  gem "image_optim_pack", "~> 0.9.1"

Downgrading Jekyll replaced sass-embedded with sassc, which builds successfully with no further tweaks.

Then, the environment built successfully, but upon running bundle exec jekyll build I got

  Could not find public_suffix-5.0.1 in any of the sources

Which was interesting to debug, to say the least. I verified that bundle env and gem environment have all the paths set correctly and public_suffix-5.0.1 is installed. After a few more hours, I added the jekyll_env derivation to $PATH, which didn't solve the issue, but made the error different.

  definition.rb:507:in `materialize': Could not find image_optim_pack-0.9.1.20230129-x86_64-linux, nokogiri-1.14.1-x86_64-linux in locally installed gems

Which is even more interesting, because now it can find public_suffix, but not nokogiri or image_optim_pack, while all 3 gems are installed in the same way.
Next morning, I decided to check Gemfile.lock once again, just to be sure that the gems are named correctly.

... They are not, not at all.

The solution was something like this:

- nokogiri (1.14.1-x86_64_linux)
+ nokogiri (1.14.1)
  ...
- image_optim_pack (0.9.1.20230129-x86_64_linux)
+ image_optim_pack (0.9.1.20230129)

Now, if you're using only Jekyll, this is the point where we reach the happy ending.

Unfortunately,

Unhelpful Error Messages And The Debugging Hell

As you might have noticed, I have some other dependencies in my Gemfile:

  • jekyll-toc is used to show the Table Of Contents you can see in the top right part of the screen if you scroll up, or at the end of this post, depending on your screen size.
  • image_optim, and by extension, image_optim_pack are used to compress images on the website before serving it.

jekyll-toc builds with no problems, but image_optim is, however, completely different:

lib/image_optim/bin_resolver/bin.rb:119:in `[]': invalid byte sequence in US-ASCII (ArgumentError)

I first tried solving this by adding locale to build inputs and setting the correct locale settings. This got me to:

lib/image_optim/bin_resolver/bin.rb:119:in `[]': invalid byte sequence in UTF-8 (ArgumentError)

Clearly missing locales is not the problem here. Digging further, I added bundle exec image_optim --info to the build script. The output was a little bit more helpful:

  ...
image_optim_pack: all bins from /nix/store/36q0biq46yha892ws5hjcnvlyzc74zsx-ruby2.7.6-image_optim_pack-0.9.1.20230129/lib/ruby/gems/2.7.0/gems/image_optim_pack-0.9.1.20230129/vendor/linux-x86_64 failed
  ...

Now, here comes the worse part: Building the same derivation locally on my desktop, with Nix on Arch Linux, worked correctly and the same command gave me this output:

  ...
image_optim_pack: all bins from /nix/store/36q0biq46yha892ws5hjcnvlyzc74zsx-ruby2.7.6-image_optim_pack-0.9.1.20230129/lib/ruby/gems/2.7.0/gems/image_optim_pack-0.9.1.20230129/vendor/linux-x86_64 worked
  ...

Somewhere around here I also tried running `nix-shell --pure` on the system that worked and realized that Jekyll (or one of the plugins?) will want a JS runtime. An easy fix: ```nix buildInputs = with pkgs; [ jekyll_env # nokogiri dependencies zlib libiconv libxml2 libxslt # jekyll wants a JS runtime nodejs-slim ]; ```

I then made sure that those directories in both environments were identical. Same executables present, same permissions, same creation date, same ownership, same everything. I made sure that $PATH is the same in both environments, I compared the outputs of bundle env and gem environment, I ran it through strace, I dumped the whole environment on both systems and diff'd them.
I did not find anything that could be meaningfully different between the 2 systems. And yet, one builds, the other does not.

That is, until I ran these commands:

sudo rm -rf /nix/store/36q0biq46yha892ws5hjcnvlyzc74zsx-ruby2.7.7-image_optim_pack-0.9.1.20230129/lib/ruby/gems/2.7.0/gems/image_optim_pack-0.9.1.20230129/vendor/darwin-x86_64
sudo rm -rf /nix/store/36q0biq46yha892ws5hjcnvlyzc74zsx-ruby2.7.7-image_optim_pack-0.9.1.20230129/lib/ruby/gems/2.7.0/gems/image_optim_pack-0.9.1.20230129/vendor/linux-i686

It builds.
It builds, successfully, correctly, with no errors.

Then, I went "oh no, this is horrible" and did nix-store --repair-path. This does not explain what the problem was, is not actually a solution, and I'm not happy with it at all, but it works... kind of.

With vendor/darwin-x86_64 and vendor/linux-i686 gone, the output of bundle exec image_optim --info is

Resolved jpegrescan 1a762f62 at /nix/store/c5giaz7fdn0xs67k4i7w62a98vg93416-ruby2.7.7-image_optim-0.31.2/lib/ruby/gems/2.7.0/gems/image_optim-0.31.2/vendor/jpegrescan
pngcrush worker: `pngcrush` not found; please provide proper binary or disable this worker (--no-pngcrush argument or `:pngcrush => false` through options)
pngout worker: `pngout` not found; please provide proper binary or disable this worker (--no-pngout argument or `:pngout => false` through options)
advpng worker: `advpng` not found; please provide proper binary or disable this worker (--no-advpng argument or `:advpng => false` through options)
optipng worker: `optipng` not found; please provide proper binary or disable this worker (--no-optipng argument or `:optipng => false` through options)
pngquant worker: `pngquant` not found; please provide proper binary or disable this worker (--no-pngquant argument or `:pngquant => false` through options)
oxipng worker: `oxipng` not found; please provide proper binary or disable this worker (--no-oxipng argument or `:oxipng => false` through options)
jhead worker: `jhead` not found, `jpegtran` not found; please provide proper binary or disable this worker (--no-jhead argument or `:jhead => false` through options)
jpegoptim worker: `jpegoptim` not found; please provide proper binary or disable this worker (--no-jpegoptim argument or `:jpegoptim => false` through options)
jpegtran worker: `jpegtran` not found; please provide proper binary or disable this worker (--no-jpegtran argument or `:jpegtran => false` through options)
gifsicle worker: `gifsicle` not found; please provide proper binary or disable this worker (--no-gifsicle argument or `:gifsicle => false` through options)
svgo worker: `svgo` not found; please provide proper binary or disable this worker (--no-svgo argument or `:svgo => false` through options)

Hey, this is better!

I fixed the rest of the missing binaries like this:

image_optim_deps = with pkgs; [
  pngout
  advancecomp
  optipng
  pngquant
  jhead
  jpegoptim
  jpeg-archive
  libjpeg
];  

Then added these dependencies to $PATH:

buildPhase = ''
  export PATH="${lib.escapeShellArg (lib.makeBinPath image_optim_deps)}":$PATH

  bundle exec jekyll build 
'';

At this point, the derivation builds successfully on both systems. Manually deleting the 2 paths is still horrible, so this is how I ended up with this Pull Request to nixpkgs:

# Clean up vendor binaries for other platforms
image_optim_pack = attrs:
  let
    vendor_path = "$out/${ruby.gemPath}/gems/image_optim_pack-${attrs.version}/vendor";
  in if builtins.currentSystem == "x86_64-linux" then {
    postInstall = ''
      rm -rf ${vendor_path}/darwin-x86_64
      rm -rf ${vendor_path}/linux-i686
    '';
  } else if builtins.currentSystem == "i686-linux" then {
    postInstall = ''
      rm -rf ${vendor_path}/darwin-x86_64
      rm -rf ${vendor_path}/linux-x86_64
    '';
  } else if builtins.currentSystem == "x86_64-darwin" then {
    postInstall = ''
      rm -rf ${vendor_path}/linux-x86_64
      rm -rf ${vendor_path}/linux-i686
    '';
} else { };

The final package I ended up with is this:

{ pkgs, lib, ... }:
let
  version = "a6476f6fc94f4476f31ebd35687dd54a807d41dd";
  repo = pkgs.fetchgit {
    url = "https://git.lain.faith/sorceress/vampysite.git";
    rev = version;
    sha256 = "1lr5jqsqik2czgas7ci4fynavhkcw6nfiw1p3lwlxm0ar3y483fk";
  };

  patched_pkgs = import (builtins.fetchTarball
    "https://github.com/AgathaSorceress/nixpkgs/tarball/image-optim-pack-cleanup") {
      config = pkgs.config;
    };

  jekyll_env = patched_pkgs.bundlerEnv {
    name = "jekyll_env";
    ruby = pkgs.ruby;

    gemdir = "${repo}/.";
  };
  image_optim_deps = with pkgs; [
    pngout
    advancecomp
    optipng
    pngquant
    jhead
    jpegoptim
    jpeg-archive
    libjpeg
  ];
in pkgs.stdenv.mkDerivation {
  inherit version;
  name = "vampysite";

  src = repo;

  buildInputs = with pkgs; [
    jekyll_env
    # nokogiri dependencies
    zlib
    libiconv
    libxml2
    libxslt
    # jekyll wants a JS runtime
    nodejs-slim
  ];

  buildPhase = ''
    export PATH="${lib.escapeShellArg (lib.makeBinPath image_optim_deps)}":$PATH

    bundle exec jekyll build 
  '';

  installPhase = ''
    mkdir -p $out
    cp -r _site/* $out/
  '';
}

Deploying

Now, the fun part. Deploying the website with Nginx is as simple as this:

# Nginx
services.nginx = {
  enable = true;

  # Use recommended settings
  recommendedGzipSettings = true;
  recommendedOptimisation = true;
  recommendedProxySettings = true;
  recommendedTlsSettings = true;

  virtualHosts."technogothic.net" = {
    useACMEHost = "technogothic.net";
    forceSSL = true;
    root = pkgs.vampysite;

    extraConfig = ''
      error_page 404 /404.html;
    '';
  };
};