Pages

Saturday, 30 July 2016

Create a static site with Hakyll, Github and Travis CI

“Static sites are fast, secure, easy to deploy, and manageable using version control.” So states the webpage for Hakyll, a great way to set up a static site or blog. It allows you to write blog posts by simply editing markdown in git, all the while having access to delicious Haskell for deeper customizations.
You can configure things to let you write blog posts directly on Github’s interface and use Travis CI to deploy your changes. Most day-to-day blogging will not require using Haskell at all or even having the Haskell environment installed on the blogger’s machine.
I’ll show you how to set everything up, including an optimized Travis config that can deploy changes in less than a minute. There is some existing information online about doing this sort of thing, but it’s all outdated in one way or another.
We’ll be using Github Pages to serve the final assets. I’ll assume you’re making a static site for a Github organization called myorg and want it to live at myorg.io.

Installation

  1. Create a Github organization. E.g. myorg
  2. Create a project myorg/myorg.github.io
  3. The master branch will be repeatedly overwritten and committed later on by Travis, so you won’t make any edits there directly. For now add a file to the root of the master branch called CNAME containing the string myorg.io
  4. Create two A records in the DNS for myorg.io pointing at 192.30.252.153 and 192.30.252.154 respectively.
  5. Generate the base Hakyll project.
    # in an empty directory of your choice
    # NOT in the git repo you've been using
    
    cabal sandbox init
    cabal install -j --disable-documentation hakyll
    cabal exec hakyll-init myorg.github.io
  6. Create an orphan source branch in your git repo and copy the generated files there.
    git checkout --orphan source
    git rm CNAME
    cp -r /path/to/generated/myorg/* .
    git add .
  7. Reuse the cabal sandbox you created earlier:
    cp -r /where/you/ran/cabal/install/.cabal-sandbox .
  8. Keep build artifacts out of git by adding these lines to .gitignore
    .cabal-sandbox
    cabal.sandbox.config
    dist/
    _cache
    _site
  9. Run your new site locally to see that it works!
    cabal sandbox init
    cabal run rebuild
    cabal run watch
    
    # now load http://localhost:8000
  10. Create .travis.yml and add the following boilerplate:
    language: haskell
    ghc: 7.8
    branches:
      only:
      - source
    before_install:
      - git submodule foreach --recursive 'git checkout master; git ls-files | grep -v README | grep -v CNAME | xargs -r git rm'
    install:
      - curl http://bin.begriffs.com/hakyll/cabal-sandbox.tar.xz | tar xJ
      - cabal sandbox init
      - cabal configure --disable-library-profiling --disable-tests --disable-library-coverage --disable-benchmarks --disable-split-objs
    before_script:
      - git config --global user.email "$GIT_EMAIL"
      - git config --global user.name "$GIT_NAME"
    script: cabal run -j build
    after_script:
      - cd _site
      - export REMOTE=$(git config remote.origin.url | sed 's/.*:\/\///')
      - git remote add github https://${GH_TOKEN}@${REMOTE}
      - git add --all
      - git status
      - git commit -m "Built by Travis ( build $TRAVIS_BUILD_NUMBER )"
      - git push github master:master | grep -v http
  11. Generate a Github auth token.
  12. Set encrypted environment variables to allow Travis to commit to the master branch
    gem install travis
    travis encrypt -r myorg/myorg.github.io --add GH_NAME="J. Doe" GH_EMAIL=jdoe@myorg.io GH_TOKEN=xxxxxxxx
  13. Commit all the files.
  14. Enable Travis for your repo. Instructions here.
  15. Push the source branch to Github.
  16. Watch the deploy progress at https://travis-ci.org/myorg/myorg.github.io
Now you can create and edit blog posts right in Github and your changes get deployed automatically.

(optional) Generating a custom cabal sandbox for Travis

You can use my shared cabal sandbox on Travis as done above, or you can build your own. It’s a little trickier. Use this Travis config as a start. It takes advantage of post-build deployment to S3.
language: haskell
ghc: 7.8
branches:
  only:
  - source
before_install:
  - travis_retry sudo add-apt-repository -y ppa:hvr/ghc
  - travis_retry sudo apt-get update
  - travis_retry sudo apt-get install --force-yes happy-1.19.3 alex-3.1.3
  - export PATH=/opt/alex/3.1.3/bin:/opt/happy/1.19.3/bin:$PATH
install:
  - cabal sandbox init
  - cabal install -j --disable-documentation -fhttps pandoc
  - cabal install -j --disable-documentation --disable-tests --reorder-goals
deploy:
  provider: s3
  access_key_id: AKIAIYJJY5B5UWSU3CFQ
  bucket: YOUR_BUCKET
  region: us-west-1
  skip_cleanup: true
  local-dir: ".cabal-sandbox"
  upload-dir: hakyll
  acl: !ruby/string:HighLine::String public_read
  on:
    repo: myorg/myorg.github.io
    branch: source
  secret_access_key:
    secure: YOUR_SECURE_KEY
 
from
https://begriffs.com/posts/2014-08-12-create-static-site-with-hakyll-github.html
-----------------------

使用 Travis 自动化部署 Hexo Blog

0x00 背景

使用 Hexo + Github Pages 搭建博客后,每次更新文章需要使用 hexo d -g 会在本地生成 public 静态博客网站和向 Github 推送的 .deploy_git 文件夹。.deploy_git 文件夹内容和 public 文件夹一致,但多了 Github 博客项目的仓库信息与提交信息。最终,.deploy_git 文件夹内的全部内容被推送到 Github 仓库中,由 Github Pages 服务完成静态网站的解析。
当切换工作环境后,需要重新安装 Nodejs 以及配置 Hexo 和它的依赖。而且每次更新文章后,都要 hexo d -g 手动部署。这样多次重复的工作非常低效,因此结合现在非常流行的 CI/CD 概念和工具,可以为 Hexo + Github Pages 博客集成 Travis CI 自动部署的能力。当推送博客仓库到 Github 后,由 Travis 自动获取当前 commit 并进行构建,把生成的静态网站推送到 Github Pages 分支。

0x01 理解 Hexo 的自动化部署

下图是 Hexo 手动部署的流程,hexo-blog 可以是本地一个项目,也可以是 Github、Gitlab 等仓库,本地配置好 Hexo 环境后,由 ① 触发部署,将本地生成的静态博客网站 .delpoy_git 推送到 Github 的静态博客仓库中。
hexo-without-travis
当引入 Travis 后,整个流程变成了下图所示的流程。hexo-blog 项目必须是一个 Github 仓库了,当有文章更新,本地由 ① 触发,把 Hexo-blog 的源码推送到 Github,剩下的工作由 Travis 完成:获取 Github hexo-blog 仓库中最新的 commit,运行我们定义的 .travis.yml 并把生成的静态博客网站 .deploy_git 推送 Github 静态博客仓库 xxx.github.io。
hexo-with-travis
注意,这里要区分 Github 中的两个仓库:静态 blog repo 和 Hexo blog repo。前者是博客网站的静态网站项目,由 Github Pages 托管和解析;后者是 Hexo 项目,前者的内容是由后者生成的。

0x02 如何配置自动化部署

再看一次引入 Travis 后的流程图,绿色箭头的流程是 Travis 自动运行的。要想实现自动化部署,需要 Travis:
  1. 能够自动获取 hexo-blog 仓库的代码提交;
  2. 能够从 hexo-blog 仓库的源码中生成静态博客网站 .deploy_git 目录;
  3. 能够把静态博客网站目录文件推送到静态博客仓库 xxx.github.io 中;
接下来,我们用三个步骤,分别解决上面提到的三个问题。

1. 配置 Travis 获取 Hexo blog

 https://travis-ci.org/account/repositories 中勾选你的 hexo-blog 项目仓库,注意,这里不是静态博客项目而是 Hexo blog 的源码仓库。
这样,每当 hexo-blog 有新的 commit 时,Travis 都能收到 Gitub 的回调,并从 hexo-blog 仓库中拉取最新提交的代码。

2. 配置 Travis 自动编译

Travis 的自动编译是由项目根目录的 .travis.yml 来控制的。我们需要为 hexo-blog 项目编写一个 .travis.yml 文件,下面是一个 .travis.yml 示例。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
language: node_js
node_js:
  - '10'

branches:
  only:
  - master

before_install:
- git config --global user.name "YOUR NAME"
- git config --global user.email "YOUR EMAIL"
- sed -i'' "s~https://github.com//.git~https://${ACCESS_TOKEN}@github.com//.git~" _config.yml

install:
- npm install hexo-cli -g
- npm install

script:
- npm run deploy
其中:
  • language  node_js 指定了使用 NodeJS docker 进行构建,NodeJS 版本为 10.x;
  • branches 指定只有 master 分支的提交才进行构建;
  • before_install 用于配置推送静态博客网站项目的 git 信息,最重要的是 sed -i 这条命令,该命令把 hexo-blog 配置文件 _config.yml 中的 git remote URL 替换为携带认证 ACCESS TOKEN 的 URL。 这样避免了 Travis 中需要输入用户名密码的问题。关于 ${ACCESS_TOKEN} 变量会在接下来进行说明;
    1
    2
    3
    4
    5
    6
    7
    # _config.yml
    ...
    deploy:
      type: git
      repo: https://github.com/ NAME>/ REPO>.git
      branch: master
    ...
  • install 指定了每次构建之前需要先安装 hexo 和 hexo-blog 项目中的依赖;
  • script 指定了运行的 npm deploy 命令去执行 hexo d -g。deploy 命令配置在了 hexo-blog 项目的 package.json 中:
    1
    2
    3
    4
    5
    6
    7
    {
      ...
      "scripts": {
        "deploy": "hexo clean && hexo d -g"
      },
      ...
    }

3. 配置 Travis 推送博客仓库的权限

Travis 需要把生成的静态博客网站推送到 Github 博客网站仓库,需要有仓库的读写权限,有两个办法为 Travis 配置权限:
  • 一个是把 Github SSH Key 放到 Hexo-blog 项目中,Travis 构建完成推送时,使用 SSH Key 推送;
  • 另一种办法是在 Github 中为 Travis 生成一个拥有读写仓库权限的 Personal access tokens。上面的 .travis.yml 中使用到的 ${ACCESS_TOKEN} 变量便是第二种方法。
关于第一种使用 Github SSH Key 的方法,因为 Hexo-blog 是公开项目,所以直接把 SSH Key 放到项目里是不安全的,因此需要使用 Travis 先把 SSH Key 加密,并在 .travis.yml 中配置解密。该方法配置稍微复杂一点,这里介绍更简单的 Personal access tokens 方法。
  1. 登录静态网站项目的 Github 账号,在 https://github.com/settings/tokens 中生成新的 token,并勾选 repo 权限。
    github-access-token
    生成的 token 一定要复制保存下来。
  2.  https://travis-ci.org/account/repositories 中点击 hexo-blog 项目的 settings,添加环境变量,把上面生成的 token 设置为变量,这里的变量名字设置为 ACCESS_TOKEN ,和 .travis.yml 中的配置一致。注意,为了安全不要勾选 DISPLAY VALUE IN BUILD LOG
    travis-env-var
至此,自动化部署已经配置好了。
在 Hexo blog 项目中新建一个 git commit,可以在 Travis 中查看项目构建的过程了,不管成功还是失败,你的邮箱都会收到一封构建邮件。

0x03 后记

不论是本文的 Travis 配置,还是其他新鲜的知识,只有理解了技术的原理,才能在遇到问题的时候知道怎么解决,看到别人的解决方法时知道他的思路和为什么要这么做,是否还有更优的方法。
---------------------------------------------------------------------------------------------------

用 Travis-CI 生成 Github Pages博客

寄宿在 Github Pages 上的静态博客通常有两种方案,其一是使用 Jekyll 方式撰写,这可以利用 Github Pages 原本就有的 Jekyll支持 生成静态网站。另一种是在 本地 也就是自己的电脑上生成好,然后把生成的 HTML 网站 push 到 Github Pages ,这种情况下 Github Pages 就完全只是一个静态页面宿主环境。

我用 Pelican 生成博客,当然就只能选择后一种方式了。这带来一些不便,比如本地配置 pelican 还是有一点点复杂的,所以不能随便找台电脑就开始写博客。有的时候只是想修正一两个错别字, 这时候必须打开某台特定的电脑才能编辑博客就显得不太方便了。再比如 pelican 本身虽然是 python 写的所以跨平台,但是具体到博客的配置方面, Windows 环境和 Linux/OSX/Unix-like 环境下还是有 些许出入 的。还有就是没有像 wordpress 那样的基于 web 的编辑环境,在手机上就不能随便写一篇博客发表出来(不知道有没有勇士尝试过在 Android 的 SL4A 环境下的 python 中跑 pelican ,还要配合一个 Android 上的 git 客户端 )。

当然并不是因此就束手无策了,感谢 Travis-CI 提供了免费的 持续整合(Continuous integration) 虚拟机环境, 通过它全自动生成静态博客成为了可能。

关于 Travis-CI

持续整合 原本是 敏捷开发(Agile Development) 或者 极限编程(Extreme Programming) 中提到的概念,大意就是说在开发的过程中, 一旦有微小的变更,就全自动地 持续 合并到主线中, 整合 变更的内容到发布版本里。 这里的 整合 实际上可以理解为 全自动测试 加上 生成最终产品 。 可以看到 持续整合 实际强调 全自动 ,于是需要有一个服务器不断地监听主线开发的变更内容, 一旦有任何变更(可以理解为 git commit )就自动调用测试和部署脚本。

于是要用持续整合就需要一个整合服务器,幸而 Travis-CI 对 github 上的公开 repo 提供了免费的整合服务器虚拟机服务,和 github 的整合非常自然。所以我们就可以用它提供的虚拟机 为博客生成静态网站。

启用 Travis-CI 自动编译

这一步很简单,访问 https://travis-ci.org/ 并用你的 Github 账户登录, 授权它访问你的账户信息就可以了。然后在 https://travis-ci.org/repositories 里开启 需要编译的 repo ,这样 Travis-CI 就会监视对这个 repo 的所有 push 操作,并且对 每个 push 调用测试了。

在 Travis-CI 中开启对 Github Repo 的持续整合

在 Travis-CI 中开启对 Github Repo 的持续整合

然后在 repo 的根目录放一个 .travis.yml 文件描述编译的步骤。 暂时 测试的目的下我写的 .travis.yml 大概是下面这样。

language: python

python:
    - "2.7"

before_install:
    - sudo apt-add-repository ppa:chris-lea/node.js -y
    - sudo apt-get update
    - sudo apt-get install nodejs ditaa doxygen parallel

install:
    - sudo pip install pelican
    - sudo pip install jinja2
    - sudo pip install babel
    - sudo pip install beautifulsoup4
    - sudo pip install markdown
    - sudo npm install -g less
    - wget "http://downloads.sourceforge.net/project/plantuml/plantuml.jar?r=&ts=1424308684&use_mirror=jaist" -O plantuml.jar
    - sudo mkdir -p /opt/plantuml
    - sudo cp plantuml.jar /opt/plantuml
    - echo "#! /bin/sh" > plantuml
    - echo 'exec java -jar /opt/plantuml/plantuml.jar "$@"' >> plantuml
    - sudo install -m 755 -D plantuml /usr/bin/plantuml
    - wget https://bintray.com/artifact/download/byvoid/opencc/opencc-1.0.2.tar.gz
    - tar xf opencc-1.0.2.tar.gz
    - cd opencc-1.0.2 && make && sudo make install && cd ..
    - sudo locale-gen zh_CN.UTF-8
    - sudo locale-gen zh_HK.UTF-8
    - sudo locale-gen en_US.UTF-8
    - sudo locale-gen ja_JP.UTF-8

script:
    - git clone --depth 1 https://github.com/farseerfc/pelican-plugins plugins
    - git clone --depth 1 https://github.com/farseerfc/pelican-bootstrap3 theme
    - mkdir output
    - env SITEURL="farseerfc.me" make publish

Travis-CI 提供的虚拟机是比较标准的 Ubuntu 12.04 LTS ,打上了最新的补丁,并且根据你指定的 语言选项会把相应的解释器和编译器升级到最新版(或者指定的版本)。这里用 python 语言的配置, 所以 python 是 2.7 的最新版并且有 pip 可以直接用。 配置中的 before_install 和 install 的区别其实不大,其中任何一个失败的话算作 build errored 而不是 build fail ,而如果在 script 里失败的话算作 build fail 。

为了编译我的模板,还需要比较新的 less.js ,所以添加了 ppa 装了个最新的 nodejs 并用它装上了 less 。 还从源码编译安装上了最新版的 opencc 1.0.2 ,因为 Ubuntu 源里的 opencc 的版本比较老(0.4), 然后 doxygen 作为 opencc 的编译依赖也装上了。 其它安装的东西么,除了 pelican 之外都是插件们需要的。以及我还需要生成 4 个语言的 locale 所以调用了 4 次 locale-gen 。由于是比较标准的 Ubuntu 环境,所以基本上编译的步骤和在本地 Linux 环境中是一样的,同样的这套配置应该可以直接用于本地 Ubuntu 下编译我的博客。

写好 .travis.yml 之后把它 push 到 github ,然后 travis 这边就会自动 clone 下来开始编译。 travis 上能看到编译的完整过程和输出,一切正常的话编译结束之后 build 的状态就会变成 passing ,比如 我的这次的build

从 Travis-CI 推往 Github

上面的测试编译通过了之后,下一步就是让 travis-ci 编译的结果自动推到 Github Pages 并发布出来。要推往 Github 自然需要设置 Github 用户的身份,在本地设置的时候是把 ssh key 添加到 github 账户就可以了,在编译细节都通过 github repo 公开了的 travis 上 当然不能放推送用的私有 key ,所以我们需要另外一种方案传递密码。

Github 上创建 Personal Access Token
Github 上创建 Personal Access Token

好在 Github 支持通过 Personal Access Token 的方式验证,这个和 App Token 一样可以随时吊销,同时完全是个人创建的。另一方面 Travis-CI 支持加密一些私密数据,通过环境变量的方式传递给编译脚本,避免公开密码这样的关键数据。

首先创建一个 Personal Access Token ,这里需要勾选一些给这个 Token 的权限,我只给予了最小的 public_repo 权限,如侧边里的图。 生成之后会得到一长串 Token 的散列码。

如果你不能使用 travis 命令
2015年2月21日更新

使用 travis encrypt 命令来加密重要数据最方便,不过如果有任何原因, 比如 ruby 版本太低或者安装不方便之类的,那么不用担心,我们直接通过 travis api 也能加密数据。

第一步用这个命令得到你的repo的 pubkey :

curl -H "Accept: application/vnd.travis-ci.2+json" https://api.travis-ci.org/repos/<github-id/repo>/key | python2 -m json.tool | grep key | sed 's/.*"key": "\(.*\)"/\1/' | xargs -0 echo -en | sed 's/ RSA//' > travis.pem

其中的 <github-id/repo> 替换成 github 上的 用户名/repo名, 比如我的是 farseerfc/farseer 。travis api 获得的结果是一个 json ,所以还用 python 的 json 模块处理了一下,然后把其中包含 key 的行用 grep 提取出来,用 sed 匹配出 key 的字符串本身,然后 xargs -0 echo -en 解释掉转义字符,然后删掉其中的 "<空格>RSA" 几个字(否则 openssl 不能读), 最后保存在名为 travis.pem 的文件里。

有了 pubkey 之后用 openssl 加密我们需要加密的东西并用 base64 编码:

echo -n 'GIT_NAME="Jiachen Yang" GIT_EMAIL=farseerfc@gmail.com GH_TOKEN=<Personal Access Token>' | openssl rsautl -encrypt -pubin -inkey travis.pem | base64 -w0

替换了相应的身份信息和token之后,这行得到的结果就是 secure 里要写的加密过的内容。

然后我们需要 travis 命令来加密这个 token , archlinux 用户可以安装 aur/​ruby-travis ,其它用户可以用 gems 安装:

$ gem install travis

装好之后,在设定了 Travis-CI 的 repo 的目录中执行一下 travis status , 命令会指导你登录 Travis-CI 并验证 repo 。正常的话会显示最新的 build 状态。 然后同样在这个 repo 目录下执行:

$ travis encrypt 'GIT_NAME="Jiachen Yang" GIT_EMAIL=farseerfc@gmail.com GH_TOKEN=<Personal Access Token>'

当然上面一行里的相应信息替换为个人的信息,作为这个命令的执行结果会得到另一长串散列码, 把这串散列写入刚才的 .travis.yml 文件:

env:
    - secure: "long secure base64 string"

有了这段声明之后, Travis-CI 就会在每次编译之前,设置上面加密的环境变量。 然后在编译脚本中利用这些环境变量来生成博客:

script:
    - git config --global user.email "$GIT_EMAIL"
    - git config --global user.name "$GIT_NAME"
    - git config --global push.default simple
    - git clone --depth 1 https://github.com/farseerfc/pelican-plugins plugins
    - git clone --depth 1 https://github.com/farseerfc/pelican-bootstrap3 theme
    - git clone --depth 1 https://$GH_TOKEN@github.com/farseerfc/farseerfc.github.io output
    - env SITEURL="farseerfc.me" make publish

after_success:
    - cd output
    - git add -A .
    - git commit -m "update from travis"
    - git push --quiet

这里要注意最后 git push 的时候一定要加上 --quiet ,因为默认不加的时候会把 代入了 $GH_TOKEN 的 URL 显示出来,从而上面的加密工作就前功尽弃了……

根据 travis 的文档 , after_success 里写的步骤只有在 script 里的全都完全无错执行完之后才会执行,这正是我们 push 的条件。目前 after_success 的成功与否不会影响到 build 的状态。 具体我用的配置见 这里的最新版 。 在我的 make github调用了 git push 命令,从而执行了 make github 之后就会自动部署到 github 上。

用 Web 编辑并发布静态博客

经过以上设置之后,一切正常的话,每次对主 repo 推送更新的同时, Travis-CI 就会自动 拉来更新然后编译并发布了。可以放置这样的图标 travisIcon 在项目的 Readme.md 中显示编译状态。

这样设置之后的另一个好处就在于可以利用 Github 的 Web 界面编辑文章内容。在 Github 里 编辑和保存之后会自动作为一个 commit 提交,所以也会触发 Travis-CI 的自动编译。

在 Github 的 Web 界面中直接编辑文章内容

在 Github 的 Web 界面中直接编辑文章内容

以及虽然目前还没有好用的 Github 的手机客户端,不过直接用 Android/iPhone 的浏览器登录 github 并编辑文章的可用性也还不错,所以同样的方式也可以直接在手机上发布博文了。

That is all, happy blogging ~

from https://web.archive.org/web/20241227234658/https://farseerfc.me/zhs/travis-push-to-github-pages-blog.html