Now also available via Tor
http://iwccx6ebuapto7dlkd4m5fooud6nqn2z7go7aass5c4q6vl6gzs5htad.onion/
Posts in category blog
Auto-deploying my blog
- Update 2020-01-09: sourcing
.bashrc
is not necessary
So, after fiddling around for several hours, I finally managed to auto-deploy my Jekyll blog from Git.
The main difficulty was that the blog is hosted on a standard shared web hoster, whereas the actual Jekyll blog content is hosted in a Git repository on a VPS.
Of course I could’ve moved the blog, which would’ve made the whole ordeal easier. But where’s the fun in that? So, after considering Weex for a moment, I reconsidered and decided to use lftp’s mirroring mode instead. A strong must was the support of FTPS, by the way. My shared hoster does allow SSH access, but I can’t (afaik) restrict access to specific directories, and I wanted to minimize impact if, god forbid, the credentials get in the wrong hands. With an FTP account, I could restrict access to a specific subdirectory, however.
My setup also uses buildbot as “CI/CD” server. It was a lot of fiddling around with the following problems:
- I use RVM to install Ruby (required for Jekyll)
and need to source the environment first (at least I think so), so I have to use a shell command of `["bash", "-c", "source ~/bash_profile && bundle install &&"]`</s> (update: this is not necessary as long as I start buildbot within this environment) - I needed to add the
haltOnFailure=True
argument to all steps (why isn’t this the default?), to make sure I don’t end up mirroring an empty directory. - I needed to create a separate worker instead of using
LocalWorker
, because only this way I could specify the umask of the worker files. The default is secure (077), which leads to the umasks being mirrored to the server by lftp, resulting in an inaccessible blog. (I don’t quite get the lftp documentation about the--no-umask
flag - what EXACTLY does it do, i.e. which umask does it use).
During my fiddling, I saw someone mentioning Drone, which might be worth a closer look sometime in the future. From what I got it supports the pipeline being set up in the repository itself, which means I wouldn’t have to fumble around with the buildbot config were I to change the “build” procedure in the future.
Also it supports docker. If I were to use a ruby docker image, I wouldn’t have this weirdness with sourcing the
.bash_profile
. But since I’m using the cheapest VPS with the lowest RAM + disk space possible, I’m quite reluctant of that option.- Update 2020-01-09: sourcing