This repository is meant to get a configuration set for installing a fresh server for Docker hosting. It’s specialized for my personal usage, but if it fits your needs, feel free to use it and give your feedback.
- Base path
- Base installation
- Some of the docker images I use in this environment
- Using docker compose
- Using Traefik v2.x as main front-end
- Back-up containers
- Clean-up FTP backups
- Rolling backups
- Using custom Docker images for Roadiz
- Rotating logs
All scripts and configurations files are written in order to perform in ~/docker-server-env
folder.
Please, adapt them if you want to clone this git repository elsewhere.
Skip this part if your hosting provider has already provisioned your server with latest docker and docker compose services.
#
# Base apps
#
sudo apt update;
sudo apt install sudo curl nano git zsh;
#
# Install oh-my-zsh
#
sh -c "$(curl -fsSL https://raw.github.com/robbyrussell/oh-my-zsh/master/tools/install.sh)"
#
# If you don’t have any password (public key only)
# Change your shell manually…
#
sudo chsh -s /bin/zsh
#
# Clone this repository in root’s home
#
git clone https://github.com/ambroisemaupate/docker-server-env.git ~/docker-server-env;
#
# Execute base installation
# It will install more lib, secure postfix and pull base docker images
#
cd ~/docker-server-env
#
# Pass DISTRIB env to install [ubuntu/debian] and EMAIL for Postfix aliases
# sudo DISTRIB="debian" EMAIL="[email protected]" bash ./install.sh if not root
sudo DISTRIB="debian" EMAIL="[email protected]" bash ./install.sh
If you are not root
user, do not forget to add your user to docker
group.
sudo usermod -aG docker myuser
sudo chown -R myuser:myuser ~/docker-server-env
Installation script install.sh
will install:
- ntp
- ntpdate
- nano
- gnupg
- htop
- curl
- zsh
- fail2ban
- postfix
- mailutils
- apt-transport-https
- ca-certificates
- software-properties-common
- clamav
- clamav-daemon
If you registered IPv6 and AAAA DNS records for your services, you must enable Docker ipv6 networking and make sure Traefik is running on a IPv6 enabled network.
Make sure to generate a unique local IPv6 range and edit etc/docker/daemon.json
before
running install.sh
script.
Check your network configuration with compose/whoami
service which prints your client information.
You can verify if IPv6 is enabled by testing if traefik is listening on both interfaces, make sure frontproxynet
is also
configured with --ipv6
option to allow traefik listening on tcp
and tcp6
:
netstat -tnlp
tcp 0 0 0.0.0.0:443 0.0.0.0:* LISTEN 2377/docker-proxy
tcp 0 0 0.0.0.0:80 0.0.0.0:* LISTEN 2398/docker-proxy
tcp6 0 0 :::443 :::* LISTEN 2383/docker-proxy
tcp6 0 0 :::80 :::* LISTEN 2404/docker-proxy
Since November 2020, hub.docker.com introduced rate limit on API, if you often pull images you'll need to setup a
Registry mirror. Install and launch compose/registry-mirror
service with your own hub.docker.com credentials in .env
.
Only use Registry mirror on private network machines or do not forget to restrict your host access to port 6000.
Copy etc/docker/daemon_with_registry.json
to your server /etc/docker/daemon.json
and restart docker
, this will setup an insecure Registry mirror on localhost:6000.
Once your Registry mirror is running on localhost:6000
you may want to use it inside your Gitlab Runners.
# /etc/gitlab-runner/config/config.toml
[runners.docker]
volumes = ["/opt/docker/daemon.json:/etc/docker/daemon.json:ro"]
Then configure /opt/docker/daemon.json
to use your host' local network IP
{
"registry-mirrors": ["http://192.168.1.xx:6000"],
"insecure-registries" : ["192.168.1.xx:6000"]
}
That way, all gitlab runners will pull Docker image through your host mirror and save precious bandwidth and rate limit.
- traefik: as the main front proxy. It handles Let’s Encrypt certificates too.
- solr (I limit heap size to 256m because we don’t usually use big document data, and it can be painful on a small VPS server)
- ambroisemaupate/ftp-backup: smart FTP/SFTP backup image
- ambroisemaupate/s3-backup: smart S3 Object Storage backup image (no need to clean-up, configure lifecycle on your S3 provider)
- ambroisemaupate/ftp-cleanup: smart FTP/SFTP backup clean-up image than delete files older than your defined limit. It won’t delete older backup files if they are the only ones available.
- ambroisemaupate/light-ssh, For SSH access directly inside your container with some useful command as
mysqldump
,git
andcomposer
. - mysql: for latest php80-alpine-nginx images and all official docker images
- gitlab-ce: If you want to setup your own Gitlab instance with a dedicated registry, all running on docker
- plausible/analytics: Awesome open-source and privacy-friendly analytics tool. Based on https://github.com/plausible/hosting.
This server environment is optimized to work with docker compose for declaring your services.
You’ll find examples to launch front-proxy and Roadiz based containers with docker compose
in compose/
folder. Just copy the sample example-se/
folder naming it with your website reference.
cp -a ./compose/example-se ./compose/mywebsite.tld
Then, use docker compose up -d --force-recreate
to create in background all your websites containers.
We need to use the same network with docker compose to be able
to discover your containers from other global containers, such as the front-proxy
and your daily backups.
See https://docs.docker.com/compose/networking/#configure-the-default-network for further details. Here is the additional lines to append to your custom docker compose applications:
networks:
frontproxynet:
external: true
Then add the frontproxynet
to your backends container that you want to expose to your front-proxy (traefik or nginx-proxy)
services:
app:
image: nginx:latest
networks:
- default
- frontproxynet
db:
image: mariadb:latest
networks:
- default
https://docs.traefik.io/providers/docker/
If install.sh
script did not setup traefik conf automatically, do:
cp ./compose/traefik/traefik.sample.toml ./compose/traefik/traefik.toml;
cp ./compose/traefik/.env.dist ./compose/traefik/.env;
touch ./compose/traefik/acme.json;
chmod 0600 ./compose/traefik/acme.json;
Then you can start traefik service with docker compose
cd ./compose/traefik;
docker compose pull && docker compose up -d --force-recreate;
Traefik dashboard will be available on a dedicated domain name: edit ./compose/traefik/.env
file to choose a monitoring host and password. We strongly encourage you to change default user and password using htpasswd -n
.
Warning: IP whitelisting won’t work correctly if you enabled AAAA (ipv6) record for your domains. Traefik won’t see
X-Real-IP
. For the moment, if you need to get correct IP address, just use ipv4.
Added backup and backup_cleanup services to your compose.yml
file:
services:
#
# AFTER your app main services (web, db, solr…)
#
backup:
image: ambroisemaupate/ftp-backup
networks:
# Container should be on same network as database
- default
depends_on:
# List here your database service
- db
environment:
LOCAL_PATH: /var/www/html
DB_USER: example
DB_HOST: db
DB_PASS: password
DB_NAME: example
FTP_PROTO: ftp
FTP_PORT: 21
FTP_HOST: ftp.server
FTP_USER: example
FTP_PASS: example
REMOTE_PATH: /home/example/backups/site
volumes:
# Populate your local path with your app service volumes
# this will backup ONLY your critical data, not your app
# code and vendor.
- private_files:/var/www/html/files:ro
- public_files:/var/www/html/web/files:ro
- gen_src:/var/www/html/app/gen-src:ro
backup_cleanup:
image: ambroisemaupate/ftp-cleanup
networks:
- default
environment:
# Make sure to use the same credentials
# as backup service
FTP_PROTO: ftp
FTP_PORT: 21
FTP_HOST: ftp.server
FTP_USER: example
FTP_PASS: example
STORE_DAYS: 5
# this path MUST exists on remote server
FTP_PATH: /home/example/backups/site
Test if your credentials are valid: docker compose run --rm --no-deps backup && docker compose run --rm --no-deps backup_cleanup
. This should launch the 2 services cleaning up older backups and
creating new ones. One for your files stored in /var/www/html
(check that you are using your main service volumes here), and a second one for your database dump.
ℹ️ You can use a .env
file in your project path to avoid typing FTP and DB credential twice.
Then add docker compose lines to your host crontab -e
(do not forget to specify your compose.yml
path):
MAILTO=""
# crontab
# You must change directory in order to access .env file
# Clean and backup "site_a" files and database at midnight
0 0 * * * cd /root/docker-server-env/compose/site_a && /usr/bin/docker compose run --rm --no-deps backup
0 1 * * * cd /root/docker-server-env/compose/site_a && /usr/bin/docker compose run --rm --no-deps backup_cleanup
# Clean and backup "site_b" files and database 15 minutes later
15 0 * * * cd /root/docker-server-env/compose/site_b && /usr/bin/docker compose run --rm --no-deps backup
15 1 * * * cd /root/docker-server-env/compose/site_b && /usr/bin/docker compose run --rm --no-deps backup_cleanup
backup_cleanup service uses a FTP/SFTP script that will check files older than $STORE_DAYS
and delete them after. It will do nothing if there are only one of each files and database backup available. This is useful to prevent deletion of non-running services by keeping at least one backup. backup_cleanup does not use sshftpfs volume to perform file listing so you can use it with every FTP/SFTP account.
Backup clean-up is already handled by your docker compose services (see above).
You can add as many backup services as you want to create rolling backups: daily, weekly, monthly:
# …
# DAILY
backup_daily:
image: ambroisemaupate/ftp-backup
depends_on:
- db
environment:
LOCAL_PATH: /var/www/html
DB_USER: test
DB_HOST: db
DB_PASS: test
DB_NAME: test
FTP_PROTO: ftp
FTP_PORT: 21
FTP_HOST: ftp.server.test
FTP_USER: test
FTP_PASS: test
REMOTE_PATH: /home/test/backups/daily
volumes:
- public_files:/var/www/html/web/files:ro
backup_cleanup_daily:
image: ambroisemaupate/ftp-cleanup
environment:
FTP_PROTO: ftp
FTP_PORT: 21
FTP_HOST: ftp.server.test
FTP_USER: test
FTP_PASS: test
STORE_DAYS: 7
FTP_PATH: /home/test/backups/daily
# WEEKLY
backup_weekly:
image: ambroisemaupate/ftp-backup
depends_on:
- db
environment:
LOCAL_PATH: /var/www/html
DB_USER: test
DB_HOST: db
DB_PASS: test
DB_NAME: test
FTP_PROTO: ftp
FTP_PORT: 21
FTP_HOST: ftp.server.test
FTP_USER: test
FTP_PASS: test
REMOTE_PATH: /home/test/backups/weekly
volumes:
- public_files:/var/www/html/web/files:ro
backup_cleanup_weekly:
image: ambroisemaupate/ftp-cleanup
environment:
FTP_PROTO: ftp
FTP_PORT: 21
FTP_HOST: ftp.server.test
FTP_USER: test
FTP_PASS: test
STORE_DAYS: 30
FTP_PATH: /home/test/backups/weekly
# MONTHLY
backup_monthly:
image: ambroisemaupate/ftp-backup
depends_on:
- db
environment:
LOCAL_PATH: /var/www/html
DB_USER: test
DB_HOST: db
DB_PASS: test
DB_NAME: test
FTP_PROTO: ftp
FTP_PORT: 21
FTP_HOST: ftp.server.test
FTP_USER: test
FTP_PASS: test
REMOTE_PATH: /home/test/backups/monthly
volumes:
- public_files:/var/www/html/web/files:ro
backup_cleanup_monthly:
image: ambroisemaupate/ftp-cleanup
environment:
FTP_PROTO: ftp
FTP_PORT: 21
FTP_HOST: ftp.server.test
FTP_USER: test
FTP_PASS: test
STORE_DAYS: 366
FTP_PATH: /home/test/backups/monthly
then launch them once a day, once a week, once a month from your crontab:
# Rolling backups (do not use same hour of night to save CPU)
# Daily
00 2 * * * cd /root/docker-server-env/compose/site_a && /usr/bin/docker compose run --rm --no-deps backup_daily
30 2 * * * cd /root/docker-server-env/compose/site_a && /usr/bin/docker compose run --rm --no-deps backup_cleanup_daily
# Weekly (on Monday early morning)
00 3 * * 1 cd /root/docker-server-env/compose/site_a && /usr/bin/docker compose run --rm --no-deps backup_weekly
30 3 * * 1 cd /root/docker-server-env/compose/site_a && /usr/bin/docker compose run --rm --no-deps backup_cleanup_weekly
# Monthly (on each 1st day)
00 4 1 * * cd /root/docker-server-env/compose/site_a && /usr/bin/docker compose run --rm --no-deps backup_monthly
30 4 1 * * cd /root/docker-server-env/compose/site_a && /usr/bin/docker compose run --rm --no-deps backup_cleanup_monthly
Example files can be found in ./compose/example-roadiz-registry/
and ./scripts/bck-example-roadiz-registry.sh.sample
if you are building custom Roadiz images with direct volumes for your websites and private registry such as Gitlab one.
Copy .env.dist
to .env
to store your secrets at one place.
After you update your website image:
docker compose pull app;
# use --no-deps to avoid recreating db and solr service too.
docker compose up -d --force-recreate --no-deps app;
# if you created a Makefile in your docker image
docker compose exec -u www-data app make cache;
Add the etc/logrotate.d/docker-server-env
configuration to your real logrotate.d
system folder.
Make sure to adapt /etc/logrotate.d/docker-server-env
file with your traefik folder location and user.
Fail2Ban is configured with a special jail traefik-auth
to block FORWARD rules for IP that trigger
too much 401 errors in ./compose/traefik/access.log
file. If you need to manually ban an IP
you must use this chain because it will prevent FORWARD rules to docker.
Make sure to edit /etc/fail2ban/jail.d/traefik.conf
with the right logpath
to Traefik access.log
file.
fail2ban-client set traefik-auth banip <IP>
You can add custom error pages to your traefik
services by adding labels to your compose.yml
file.
All html
files are stored in compose/traefik/service-error/html
folder, and served by Nginx in traefik-service-error
service.
Behind the scene, it is an Nginx docker container running with a custom compose/traefik/service-error/default.conf
configuration:
all requests except for /css
, /img
are redirected to /404.html
or /503.html
files.
You can use a custom folder by changing volume path in compose.yml
file:
volumes:
- ./service-error/html:/usr/share/nginx/html:ro
- ./service-error/default.conf:/etc/nginx/conf.d/default.conf:ro
labels:
# Custom error pages
- "traefik.http.middlewares.${APP_NAMESPACE}_errors.errors.status=500-599"
- "traefik.http.middlewares.${APP_NAMESPACE}_errors.errors.service=traefik-service-error-traefik"
- "traefik.http.middlewares.${APP_NAMESPACE}_errors.errors.query=/{status}.html"
Traefik is configured to serve a catch-all error page for all other errors and non-existing services.
It will serve compose/traefik/service-error/503.html
file.
You can change the catch-all behaviour in compose/traefik/compose.yml
file by editing traefik-service-error
service labels.
labels:
- "traefik.enable=true"
# Serve catch-all error pages on HTTP
- "traefik.http.routers.traefik-service-error-traefik.priority=1"
- "traefik.http.routers.traefik-service-error-traefik.rule=HostRegexp(`{host:.+}`)"
- "traefik.http.routers.traefik-service-error-traefik.entrypoints=http"
# Serve catch-all error pages on HTTPS
- "traefik.http.routers.traefik-service-error-traefik-secure.priority=1"
- "traefik.http.routers.traefik-service-error-traefik-secure.rule=HostRegexp(`{host:.+}`)"
- "traefik.http.routers.traefik-service-error-traefik-secure.entrypoints=https"
- "traefik.http.routers.traefik-service-error-traefik-secure.tls=true"
- "traefik.http.routers.traefik-service-error-traefik-secure.tls.certresolver=letsencrypt"
If you are hosting multiple database servers on the same server using Docker, you may want to increase the number of
fs.aio-max-nr
to avoid EAGAIN
errors.
# Check current value
sudo cat /proc/sys/fs/aio-max-nr
# Set new value (not persistent)
sudo sysctl -w fs.aio-max-nr=200000
If you want this value to be persisted, you can add it in /etc/sysctl.conf
or any /etc/sysctl.d/*.conf
file.