Migrate to the Pixelfed + Docker Compose v2
There are a lot of changes in how Pixelfed Docker/Docker Compose images work - it's a complete rewrite - with a couple of breaking/significant changes.
But don't worry! This document and the Migration Guide further down covers all of the required changes and steps you need to safely migrate.
We don't take breaking changes or lengthy migration guides lightly, and future iterations and improvements will be either seamless or much smaller in scope and complexity.
We do however believe that the many improvements (listed below) are worth it, and sets us up for a brighter future for self-hosted and maintainable Pixelfed servers
Breaking Changes
These changes are breaking, removing, or changing existing behavior from Docker Compose v1 and requires your attention and possibly some steps to fix.
The Migration Guide has step by step guide for most of these!
New Dockerfile
Breaking
All runtime variations of Pixelfed on Docker are now build from a single Dockerfile
, rather than the previously three.
Please see the Docker Runtimes and Docker Customization documentation for more information.
New Docker Compose file Breaking
The docker-compose.yml
file has been rewritten from the ground up to be provide better defaults and more flexible out of the box.
This mean that most things you would likely want to tweak can be controlled from your .env
config file, like installing additional APT Packages, PECL/PHP extensions, changing PHP version, modifying PHP settings (such as memory_limit
), disabling specific services, changing where data and config files are stored, and so on.
New .env
/ .env.docker
file Breaking
The starter .env.docker
file has been significantly expanded with most of the Pixelfed settings available, along with many Docker Compose specific ones.
Your Pixelfed specific settings are (of course) unchanged, but you need to configure them again after copying the new .env.docker
file.
Using bind
volumes Breaking
The old docker-compose.yml
configuration file declared four anonymous Docker volumes for storing Pixelfed related data within.
These are no longer used, instead favoring a Docker bind volume approach where content is stored directly on the server disk, outside of a Docker volume.
The consequence of this change is that all data stored in the - now unsupported - Docker volumes will no longer be accessible by Pixelfed.
- The
db-data
volume definitely contains important data - it's your database, after all! - The
app-storage
volume definitely contains important data - it's files uploaded to - or seen by - your server! - The
redis-data
volume might contain important data (depending on your configuration) - The
app-bootstrap
volume does not contain any important data - all of it will be generated automatically in the new setup on startup. We will not be migrating this!
Please see the Migration steps section for information on moving your data.
No Docker networks Breaking change
The docker-compose.yml
no longer include any custom networks, instead favoring the simplicity of the default bridge
network.
The published ports can be configured via your .env
file, look for the DOCKER_*_HOST_PORT
keys.
Changes needing attention
These changes are mostly informational and unlikely to cause any issues during upgrade, but included for visibility and in case your specific set up require them to be changed.
New Redis version Attention needed
Redis has been upgraded from using the last version 5
release to using the latest 7.2
version instead - favoring the debian
variant instead of alpine
.
The Redis version can be controlled in your .env
file via DOCKER_REDIS_VERSION
.
If you want to keep using the Redis version set DOCKER_REDIS_VERSION="5-alpine"
in your .env
file.
If you want to use 7.2
but the alpine variant, simply append -alpine
to the DOCKER_REDIS_VERSION
(e.g. DOCKER_REDIS_VERSION=7.2-alpine
).
Pinned MariaDB version Attention needed
The included db
service is now pinned to MariaDB 11.2
instead of latest
.
You can revert to the previous setting by setting DB_VERSION="latest"
in your .env
file.
Automatic "One-time setup tasks" Attention needed
This is also covered in the migration guide but existing Pixelfed instances need to disable the automatic run of One-time setup tasks by setting DOCKER_APP_RUN_ONE_TIME_SETUP_TASKS=0
in your .env
file.
Your container will fail to start up if these are already configured for your site.
New or improved features
All the new and exciting features and capabilities. 🚀
This is where we hope we can convince you that the breaking changes and migration work was worth it ❤️
Nginx Proxy service New
The new docker-compose.yml
includes an optional (but enabled by default) Nginx Proxy for SSL/TLS termination.
Please see How do I use my own Proxy server?
and DOCKER_PROXY_*
keys in the .env
file for more information.
LetsEncrypt/ACME service New
The new docker-compose.yml
includes an optional (but enabled by default) ACME/LetsEncrypt service that when combined with the new Nginx Proxy automatically creates and maintains your SSL/TLS certificates.
Please see How do I use my own SSL certificate?
and DOCKER_PROXY_*
+ LETSENCRYPT_*
keys in the .env
file for more information.
Automatic run of "One-time setup tasks" New
When you set up a new Pixelfed server, there is handful of commands you need to run once (and only once) - these are called One-time setup tasks.
These steps are now automatically run for you when creating a new Pixelfed server.
Extensive documentation New
How to run and use the new Docker setup has been documented extensively. Not only on this page, but also in all the new scripts, Dockerfile
, and .env.docker
file.
We hope this will make it much easier to confidently and comfortably run your Pixelfed server with Docker - and when things do go wrong, the debugging and fixing of the issue much easier and quicker.
Automatic publishing of Docker images Improved
Going forward we will automatically build and push Docker images for Pixelfed to both Docker Hub and GitHub Container Registry.
We will be automatically pushing all combinations of our supported Docker runtimes.
Further more, testing Pull Requests has never been easier, as we will also build and push Docker images for all Pull Requests with tag prefix {RUNTIME}-pr-{ID}
(e.x. gcr.io/pixelfed-glitch/pixelfed:apache-pr-4844
).
You can control which Pixelfed release you use via the DOCKER_APP_RELEASE
key/value pair in your .env
file.
Customizable Dockerfile
New
It's now possible to tweak many Docker related settings without copying or forking Pixelfed Dockerfile!
The new Dockerfile has many Build Arguments (e.g., --build-arg
) allowing you to easily
- Change your PHP version
- Install extra PECL extensions
- Install extra PHP extensions
- Install extra APT packages
- Change Debian version
All of these (and more) settings are controlled from your .env
file, and you simply need to run docker compose build
to build your own bespoke version of Pixelfed on your server!
Customizable ENTRYPOINT
New
Do you want to run a script on container start up? Or perhaps to disable a specific init script? Or even disable all of the init scripts?
Now you can! And like with all the other settings it's just a setting in your .env
file!
Templating files New
The new ENTRYPOINT
system offers an extensible and flexible way for you to template configuration files such as your php.ini
during container start up.
The templating system has access to all settings from your .env
file, and you can easily add your own templates!
Automatic PHP/Web server configuration New
Thanks to the new templating system we now automatically calculate and configure required PHP / Web server settings for you - using your .env
settings - such as
- (php.ini)
upload_max_filesize
with this formula(MAX_PHOTO_SIZE * MAX_ALBUM_LENGTH) + BUFFER
- (php.ini)
post_max_size
with this formula(MAX_PHOTO_SIZE * MAX_ALBUM_LENGTH) + BUFFER
- (php.ini)
max_file_uploads
usingMAX_ALBUM_LENGTH
- (php.ini)
memory_limit
usingPHP_MEMORY_LIMIT
- (php.ini)
[Date]date.timezone
usingAPP_TIMEZONE
Permission auto-fixing New
If your installation for some reason has issues with permissions to files/directories, you can now automatically fix ownership and permissions on start up.
This is an opt-in feature.
Faster Docker image building Improved
We now utilize BuildKit, layer caching, multi-stage, and multi-platform to really speed up the building of Docker images.
Under ideal conditions, a docker compose build
can now complete in less than a minute for most PHP changes. This of course also improve the speed of building Docker images in GitHub Actions and CI!
Fork friendly Docker releasing New
Forks of Pixelfed Glitch that have enabled GitHub Actions should automatically have the same Docker build + push experience as pixelfed-glitch/pixelfed have.
The docker
workflow will by default build + push to the GitHub Container Registry for the project the commit was made to, meaning any downstream projects will not have to fork or modify their docker
workflow to have a reliable release process.
Forks can further more set GitHub Actions Project variables DOCKER_HUB_USERNAME
, DOCKER_HUB_ORGANISATION
, DOCKER_HUB_REPO
and DOCKER_HUB_TOKEN
for automatic pushing of images to Docker Hub as well. Please see the .github/workflows/docker.yml
file for more information.
Migration steps
INFO
This is a best-effort guide to help migrate off the old system. The operation is potentially complicated (and risky), so please be careful!
DANGER
PLEASE MAKE SURE TO BACKUP YOUR SERVER AND DATA BEFORE ATTEMPTING A MIGRATION
YOUR INSTANCE WILL BE DOWN WHILE DOING THE MIGRATION; PLEASE PLAN ACCORDINGLY; DEPENDING ON DATA SIZE IT COULD TAKE ANYWHERE FROM 5 MINUTES TO 5 HOURS
1) Backup
- Make sure to back up your server (ideally after step 1 below has been completed, but before is better than not at all!)
- Capture the current Git version / Pixelfed release you are on (e.g.,
git --no-pager log -1
outputs the commit reference as the 2nd word in the first line) - Backup your
.env
file (we will do this in step 3 as well) - Backup your
docker-compose.yml
file (cp docker-compose.yml docker-compose.yml.old
) - Read through the entire document before starting
2) Migrate .env
file
The new .env
file for Docker is a bit different from the old one (many new settings!) so the easiest is to grab the new .env.docker
file and modify it from scratch again.
cp .env .env.old
wget -O .env.new https://raw.githubusercontent.com/pixelfed-glitch/pixelfed/main/.env.docker
Then open your old .env.old
configuration file, and for each of the key/value pairs within it, find and update the key in the new .env.new
configuration file.
Don't worry, though; the file might look different (and significantly larger), but it behaves exactly the way the old file did; it just has way more options!
TIP
Don't worry if a key is missing in .env.new
; you can add those key/value pairs back to the new file - ideally in the Other configuration
section near the end of the file - but anywhere should be fine.
This is a great time to review your settings and familiarize yourself with all the new ones.
INFO
In particular the following sections
PHP configuration
section (near the end of the file) where- The
DOCKER_APP_PHP_VERSION
settings control your PHP version - The
PHP_MEMORY_LIMIT
settings control your PHP memory limit
- The
Docker Specific configuration
section (near the end of the file) where- The
DOCKER_ALL_HOST_DATA_ROOT_PATH
setting dictates where the newly migrated data will live. - The
DOCKER_APP_RUN_ONE_TIME_SETUP_TASKS
controls whether theOne-time setup tasks
should run. We do not want this since your Pixelfed instance is already set up!
- The
- Frequently Asked Question / FAQ
3) Stop containers
WARNING
This will take your Pixelfed instance offline
Stop all running containers (web, worker, redis, db)
docker compose down
4) Update source code
Update your project to the latest release of Pixelfed by running:
git pull origin $release
INFO
The $release
can be any valid git reference like develop
, staging
, latest
, or a tagged release such as v0.12.0-glitch.1.4.0
.
5) Migrate data
The migration guide temporarily branches into two paths here:
Which Docker volume type am I using?
If your old docker-compose.yml
had a volumes
configuration section like below, you used should follow Path A.
This is also what the v1 docker-compose.yml
file for pixelfed/pixelfed used.
volumes:
db-data:
redis-data:
app-storage:
app-bootstrap:
A) Anonymous volumes
WARNING
It's important to note that this is a copy operation - so disk usage will (temporarily) double while you migrate
We provide a "migration container" for your convenience that can access both the new and old volumes, allowing you to copy the data into the setup.
You can use mv
(move files) instead of rsync
in the guide below - but this is a be a destructive action, so you can't quickly roll back to your old setup, as the data no longer exists in the anonymous Docker volumes.
A1) Run migration container
You can access the Docker container with both old and new volumes by running the following command:
docker compose -f docker-compose.migrate.yml run migrate bash
This will put you in the /migrate
directory within the container, containing 9 directories like shown here:
|-- app-storage
| |-- new
| `-- old
|-- db-data
| |-- new
| `-- old
`-- redis-data
|-- new
`-- old
A2) Check old folders
First thing we want to do is to check if the data inside the container looks correct!
The following commands should all return SOME files and data - if they do not - then there might be an issue with the anonymous volume binding.
INFO
The content of the old folders may not be exactly whats show in the examples below - if a couple of the files or folders match, it's extremely likely it's correct!
$ ls app-storage/old
app debugbar docker framework logs
oauth-private.key oauth-public.key purify
# Redis data might also be entirely empty, thats *okay*
$ ls redis-data/old
appendonlydir server.pid
$ ls db-data/old
aria_log_control ddl_recovery-backup.log ib_buffer_pool ib_logfile0
ibdata1 mariadb_upgrade_info multi-master.info mysql
performance_schema pixelfed_prod sys undo001 undo002 undo003
A3) Check new folders
The following commands should all return NO files and data - if they contain data - you need to delete it (backup first!) or skip that migration step.
If you haven't run docker compose up
since you updated your project in step (2) - they should be empty and good to go.
ls app-storage/new
ls db-data/new
ls redis-data/new
A4) Copy data
WARNING
This is where we potentially will double your disk usage (temporarily)
Now we will copy the data from the old volumes to the new ones.
The migration container has rsync
installed - which is perfect for that kind of work!
NOTE It's important that the "source" (first path in the rsync
command) has a trailing /
- otherwise, the directory layout will turn out wrong!
NOTE Depending on your server, these commands might take some time to finish; each command should provide a progress bar with a rough time estimation.
NOTE rsync
should preserve ownership, permissions, and symlinks correctly for you and all the files copied.
Let's copy the data by running the following commands:
rsync -avP app-storage/old/ app-storage/new
rsync -avP db-data/old/ db-data/new
rsync -avP redis-data/old/ redis-data/new
A5) Sanity checking
Let's make sure everything is copied over successfully!
Each new directory should contain something like (but not always exactly) the following - NO directory should have a single folder called old
; if they do, the rsync
commands above didn't work correctly - and you need to move the content of the old
folder into the "root" of the new
folder like shown a bit in the following sections.
The redis-data/new directory might also contain a server.pid
$ ls redis-data/new
appendonlydir
The app-storage/new directory should look something like this
$ ls app-storage/new
app debugbar docker framework logs oauth-private.key oauth-public.key purify
The db-data/new directory should look something like this. There might be a lot of files or very few files, but there must be a mysql
, performance_schema
, and ${DB_DATABASE}
(e.g., pixelfed_prod
directory)
$ ls db-data/new
aria_log_control ddl_recovery-backup.log ib_buffer_pool ib_logfile0 ibdata1 mariadb_upgrade_info multi-master.info mysql performance_schema pixelfed_prod sys undo001 undo002 undo003
If everything looks good, type exit
to leave the migration container.
B) Bind/Host volumes
If you used Bind/Host volumes, then this guide can't offer any hard step-by-step guide to move your data, but instead will let you know how things would look if you started from scratch, so you can mirror or change this behavior.
In your .env
file, the following KEY/VALUE pairs controls where your data and config files will go - please review each and make sure your files are in these paths, or adjust them to fit your current layout.
You can see in the docker-compose.yml
file for each service volume
section how they are used.
DOCKER_ALL_HOST_DATA_ROOT_PATH
- The "root" path for all service data.Default:
./docker-compose-state/data
DOCKER_ALL_HOST_CONFIG_ROOT_PATH
- The "root" path for all service configs.Default:
./docker-compose-state/config
DOCKER_APP_HOST_STORAGE_PATH
(/var/www/storage
) - Path for Pixelfedstorage
, e.x., uploads by users, emojis, and suchDefault:
${DOCKER_ALL_HOST_DATA_ROOT_PATH}/pixelfed/storage
DOCKER_APP_HOST_CACHE_PATH
(/var/www/bootstrap/cache
)Path for Pixelfed
bootstrap/cache
data.Default:
${DOCKER_ALL_HOST_DATA_ROOT_PATH}/pixelfed/cache
DOCKER_REDIS_HOST_DATA_PATH
Path where
Redis
will store it's data.Default:
${DOCKER_ALL_HOST_DATA_ROOT_PATH}/redis
DOCKER_DB_HOST_DATA_PATH
Path where your database (
MariaDB
) store it's data.Default:
${DOCKER_ALL_HOST_DATA_ROOT_PATH}/db
6) Start containers
With all an updated Pixelfed (step 2), updated .env
file (step 3), and migrated data (steps 4, 5, 6, and 7), we're ready to start things back up again.
But before we start your Pixelfed server, let's put the new .env
file we made in step 1 in its proper place.
cp .env.new .env
The Database
The first thing we want to try is to start up the database by running the following command and checking the logs:
docker compose up -d db
docker compose logs --tail 250 --follow db
If there are no errors and the server isn't crashing, great! If you have an easy way of connecting to the database via a GUI or CLI client, do that as well and verify the database and tables are all there.
Redis
The next thing we want to try is to start up the Redis server by running the following command and checking the logs:
docker compose up -d redis
docker compose logs --tail 250 --follow redis
if there are no errors and the server isn't crashing, great!
Worker
The next thing we want to try is to start up the Worker server by running the following command and checking the logs:
docker compose up -d worker
docker compose logs --tail 250 --follow worker
The container should output a lot of logs from the docker-entrypoint system, but eventually you should see these messages
Configuration complete; ready for start up
Horizon started successfully.
If you see one or both of those messages, the worker seems to be running.
If the worker is crash looping, inspect the logs and try to resolve the issues.
You can consider the following additional steps:
- Enabling
DOCKER_APP_ENTRYPOINT_DEBUG
, which will show even more log output to help understand what is going on - Enabling
DOCKER_APP_ENSURE_OWNERSHIP_PATHS
against the path(s) that might have permission issues - Fixing permission issues directly on the host since your data should all be in the
${DOCKER_ALL_HOST_DATA_ROOT_PATH}
folder (./docker-compose-state/data
by default)
Web
The final service, web,
will bring your site back online! What a journey it has been.
Let's get to it: run these commands to start the web
service and inspect the logs.
docker compose up -d web
docker compose logs --tail 250 --follow web
The output should be identical to that of the worker
, so please see that section for debugging tips if the container is crash looping.
If the web
service comes online without issues, start the rest of the (optional) services, such as the proxy
, if enabled, by running:
docker compose up -d
docker compose logs --tail 250 --follow
If you changed anything in the .env
file while debugging, some containers might restart now; that's perfectly fine.
7) Verify
With all services online, it's time to go to your browser and check everything is working.
- Upload and post a picture
- Comment on a post
- Like a post
- Check Horizon (
https://${APP_DOMAIN}/horizon
) for any errors - Check the Docker compose logs via
docker compose logs --follow
If everything looks fine, yay, you made it to the end! Let us do some cleanup
8) Cleanup
With everything working, please take a new snapshot/backup of your server before we do any cleanup. A post-migration snapshot is handy since it contains both the old and new configuration + data, making any recovery much easier in a rollback scenario later.
Now, with all the data in the new folders, you can delete the old Docker Container volumes (if you want, completely optional)
List all volumes and give them a look:
docker volume ls
The volumes we want to delete end with the volume name (db-data
, app-storage
, redis-data
, and app-bootstrap
.) but have some prefixes in front of them.
Once you have found the volumes in in the list, delete each of them by running:
docker volume rm $volume_name_in_column_two_of_the_output
You can also delete the docker-compose.yml.old
and .env.old
files since they are no longer needed:
rm docker-compose.yml.old
rm .env.old
Rollback
Oh no, something went wrong? No worries; you have backups and a quick way back!
Move docker-compose.yml
back
cp docker-compose.yml docker-compose.yml.new
cp docker-compose.yml.old docker-compose.yml
Move .env
file back
cp env.old .env
Go back to the old source code version
git checkout $commit_id_from_step_0
Start things back up
docker compose up -d
Verify it worked
See Step 7 for recommended steps to verify everythin is working