In this post I’ll break down the setup of my self-hosted virtual home: https://s-ol.nu.
First a quick overview of what this guide will cover:
- HTTPS server with multiple subdomains and varying backends
- traefik reverse-proxy maintains SSL certificates and serves all requests
- docker-compose manages running sites/microservices
- a private-public git server
- micro ‘CI’ setup rebuilds & redeploys docker images when updates are pushed
UPDATE (2019-10-03): I updated the git hook below to one that supports pushing and building
multiple branches based on the docker-compose.yml
.
Most of these projects are very well documented so I won’t go into a lot of detail on setting them up.
HTTPS Server
To run multiple subdomains from a single machine, the HTTP requests need to be matched according to the ‘Host’ header. Most HTTP servers have good facilities for doing this, e.g. apache has vhosts, nginx has server directives etc.
In the past I had used apache as my main server, which worked well for static content and PHP apps, but not all applications fit into this scheme too well and configuration is a tad tedious.
With this latest iteration I am using traefik as a “reverse proxy”. This means traefik doesn’t serve anything (not even static content) by itself, it just delegates requests to one of multiple configured services based on a system of rules.
It also handles letsencrypt certificate generation and updates out-of-the-box and can be tightly integrated with docker, so that it automatically reacts to new services being added.
Traefik can be run at system level, but currently I prefer installing the least system-level applications to have my setup as self-contained as possible. Therefore I went with this traefik in docker-compose setup from kilian.io:
version: '3.4'
services:
traefik:
image: traefik:1.5-alpine
restart: always
ports:
- 80:80
- 443:443
networks:
- web
volumes:
- /var/run/docker.sock:/var/run/docker.sock:ro
- ./traefik.toml:/traefik.toml
- ./acme.json:/acme.json
container_name: traefik
networks:
web:
external: true
with the small addition of the :ro
at the end of the docker socket volume,
to prevent attacks on traefik from being able to take over the host docker system (too easily).
In the guide you can find more details, including the traefik.toml
that I am using almost verbatim.
Hosting Sites
With traefik set up, dockerized services can be added and exposed trivially.
For example to start redirectly
, a tiny link redirect service, this addition suffices:
redirectly:
image: local/redirectly:master
restart: always
networks:
- web
labels:
- "traefik.frontend.rule=Host:s-ol.nu"
- "traefik.enable=true"
By setting different subdomains in the frontend-rule section, many different services can be provided.
The image local/redirectly:git
in this case is built automatically when a repo is pushed (see below).
note: if a container doesn’t have an EXPOSE directive, or EXPOSEs multiple ports,
you will have to add a traefik.port
label specifying which port to use.
private/public Git Server
While I still have a lot of code on Github, where collaboration is easy, I prefer to own the infrastructure that I store my private projects on. I also wanted to have a public web index of some of the projects.
The git infrastructure itself is mananged by gitolite, which I really enjoy using.
repo gitolite-admin @all
RW+ = s-ol
C = s-ol
repo public/.*
R = @all daemon
option writer-is-owner = 1
repo ... ...
RW+ = ludopium
In the first block I grant myself full access to all repos, as well as the right to automatically create repos by attempting to push/pull from them.
The second block makes all repos prefixed with public/
readable by anyone in the gitolite system,
as well as the git-daemon
, which allows cloning via git://....
access (port 9418).
The write-is-owner
option lets me set the git description
field using ssh git@git.s-ol.nu desc
.
I chose the public/
prefix because it results in all public repos being stored in one directory together
(/var/lib/gitolite/repositories/public
), where klaus can easily pick them up.
The klaus web frontend is set up using traefik above like so:
klaus:
image: hiciu/klaus-dockerfile
restart: always
networks:
- web
volumes:
- /var/lib/gitolite/repositories/public:/srv/git:ro
command: /opt/klaus/venv/bin/uwsgi --wsgi-file /opt/klaus/venv/local/lib/python2.7/site-packages/klaus/contrib/wsgi_autoreload.py --http 0.0.0.0:8080 --processes 1 --threads 2
environment:
KLAUS_REPOS_ROOT: /srv/git
KLAUS_SITE_NAME: git.s-ol.nu
KAUS_CTAGS_POLICY: tags-and-branches
KLAUS_USE_SMARTHTTP: y
labels:
- "traefik.frontend.rule=Host:git.s-ol.nu"
- "traefik.enable=true"
I am using the ‘autoreload’ feature and the hiciu/klaus-dockerfile
docker image.
Setting KLAUS_USE_SMARTHTTP
allows cloning repos via HTTP.
In the future I would like to modify klaus a bit, for example by showing the README in the root of a project per default and applying a custom theme.
Micro-CI
The last piece of the puzzle is automatically deploying projects whenever they are pushed.
This can be realized using git’s post-receive
hooks and is generally pretty well known.
I followed this gitolite guide for storing repo-specific hooks in the gitolite-admin repo. It requires a change in the gitolite rc file (on the server), but after that you can configure deployment processes in the conf like this:
@dockerize = public/redirectly ...
@jekyllify = blog
repo @dockerize
option hook.post-receive = docker-deploy
# i actually dont have a jekyll blog anymore but its an easy one as well
repo @jekyllify
option hook.post-receive = jekyll-deploy
The hooks are stored in the same repo under local/hooks/repo-specific
.
Here is the docker-deploy
hook I am using:
#!/bin/bash
set -e
while read oldrev newrev refname
do
BRANCH="$(git rev-parse --symbolic --abbrev-ref $refname)"
# Get project name
PROJECT="$PWD"
PROJECT="${PROJECT#*/repositories/public/}"
PROJECT="${PROJECT#*/repositories/}"
PROJECT="${PROJECT%.git}"
PROJECT="$(echo "$PROJECT" | tr "/ " "-_")"
# Paths
CHECKOUT_DIR=/tmp/git/$PROJECT
TARGET_DIR=/home/s-ol/aerol
IMAGE_NAME=local/$PROJECT:$BRANCH
# this one doesn't require python & yq, but it means the container has to run already...
# SERVICES=$(docker ps --filter "ancestor=${IMAGE_NAME}" --format '{{.Label "com.docker.compose.service"}}' \
# | sort | uniq)
SERVICES=$(yq -r <"$TARGET_DIR/docker-compose.yml" \
".services | to_entries | map(select(.value.image == \"${IMAGE_NAME}\").key) \
| join(\" \")")
if [ -z "$SERVICES" ]; then
continue
fi
mkdir -p "$CHECKOUT_DIR"
GIT_WORK_TREE="$CHECKOUT_DIR" git checkout -q -f $newrev
echo -e "\e[1;32mChecked out '$PROJECT'.\e[00m"
cd "$CHECKOUT_DIR"
docker build -t "$IMAGE_NAME" .
echo -e "\e[1;32mImage '$IMAGE_NAME' built.\e[00m"
cd "$TARGET_DIR"
docker-compose up -d $SERVICES
echo -e "\e[1;32mService(s) '$SERVICES' restarted.\e[00m"
done
It will build a local/$REPO:$BRANCH
image whenever you push, then run docker-compose up -d $SERVICES
in $TARGET_DIR
,
where $SERVICES
are all the docker-compose services that use the image. If there are none, no image will be built.
For this to work it has to parse the docker-compose.yaml
file, which means you have to install yq
and jq
, e.g. on Ubuntu:
sudo apt-get install jq python3-pip
sudo pip install yq
If you would like to avoid that, you can use the commented command for SERVICES=
above, which only relies on docker itself,
the only problem is that you will have to do the first build manually (or re-tag a dummy image) before the first build,
since it can only detect containers that are already running.
That’s basically it! If you have questions or comments i’ll be happy to hear from you on twitter, github or mastodon.