Dockerfile update

[hjhornbeck]

While Docker isn't officially supported by Hometown, leaving the Mastodon 3.5.5 Docker configuration in place with the new 4.0.2 code is a bad idea. At minimum, you'll have a stale Node install that's months behind on security updates. There are some minor tweaks to the default configuration, but they're flagged by comments so they're easy to revert or modify as necessary.

Running Hometown on Docker

I'll by typing up my own longer blog post in due time, but there's no harm dropping a cheat sheet here. By following this outline, I was able to upgrade a Hometown 1.0.8 install to 1.1.0 with nothing worse than a minute or two of downtime.

My configuration uses the GitHub repository as its source, rather than images drawn from DockerHub. I like to tweak and fiddle with my setup, especially the themes, and I'm happy to sacrifice some disk space for the privilege.

Installing from Scratch

This is by far the easiest approach, you just follow one of the existing guides for running Mastodon via Docker, pause after you've set up .env.production, add any Hometown-specific features to it as per the Wiki, then resume what the guide says to do.

If you're enabling ElastiSearch, the second of the two guides has some additional actions you'll need to do, plus be aware of this bug in Mastodon which can quietly block ES from working at all.

Upgrading from Hometown 1.0.8

Here's how I accomplished this. I committed any leftover changes, then ran these commands from the non-Docker instructions in the root of my local Hometown repository:

git remote update
git checkout v4.0.2+hometown-1.1.0

This "wiped out" my customizations, but as I committed them all to a branch I can reconstruct them later via diffs. I then ran:

sudo docker-compose build

to build the new image. The old image will continue running in the background, as per usual. I like adding 2>&1 | less to the end and mashing PgDn, as if a compilation error happens it almost invariably requires scrolling back a few screens to find the issue.

If the build succeeded, we're almost clear to start the dangerous portion. If you're running on the cloud, now would be a great time to take a snapshot. Whatever the case, you should back up the existing database. If you haven't changed the defaults from the Dockerfile, then

sudo docker exec -it hometown_db_1 pg_dump -U postgres -Fc postgres > hometown.db.dump

should do the trick. If you have changed the defaults, you may need to use sudo docker ps to figure out the name of the PostgreSQL image to swap in place of "hometown_db_1", then browse through .env.production to extract the username to place after -U and the database name to place after -Fc. The Hometown docs don't say how to restore the database should the process go South, but after reading a manpage or two I think the magic words are roughly

sudo docker exec -it hometown_db_1 pg_restore -U postgres --clean --if-exists -d postgres < hometown.db.dump

Now we're ready for the scary "you could destroy everything" part. All the earlier commands are trivial to roll back, but after this point any delay could cause data corruption. As per the Hometown docs, run the pre-deployment database migrations.

sudo docker-compose run -e SKIP_POST_DEPLOYMENT_MIGRATIONS=true -e RAILS_ENV=production --rm web bundle exec rails db:migrate

where web is the name of the webserver image in docker-compose.yml. The docs state you should precompile all assets next, but I'm 95% sure they were already built when you ran sudo docker-compose build. If you're paranoid and want to be absolutely sure precompilation is done, then at this stage run:

sudo docker-compose run -e RAILS_ENV=production --rm web bundle exec rails assets:precompile

Here, the Hometown docs say you should run the post-deployment migrations. In Docker-ese:

sudo docker-compose run -e RAILS_ENV=production --rm web bundle exec rails db:migrate

Finally, we need to stop the old images and spin up the new ones. Run:

sudo docker-compose up -d

and give Docker some time to finish rotating. A quick sudo docker ps should confirm the new images are booting up, and in a short while (10-15 seconds for the teeny-tiny instance I manage) you should be back to fully functional.

[dariusk]

Merged! @hjhornbeck, could I perhaps add the documentation here to the wiki? Or at least link to it (since you have write access here but not to the wiki).

[hjhornbeck]

Sure, feel free to add this to the wiki! That blog post I mentioned will still be a while, and I don't think it'll add much.

[hjhornbeck]

One and a half years later, I've re-used the same basic recipe to install the 4.2-1.1.1 security update. Very little has changed, for instance the preferred method to invoke Docker Compose is now docker compose instead of docker-compose. I also should have mentioned pre-compiling the assets before running the pre-deployment database scripts, as in that rare 5% chance you're not stuck waiting for the compile to finish before you can bring the new images up.

I did find a better way to back up the database, however:

sudo docker compose exec db pg_dump -U postgres -Fc postgres > ~/backup.dump

Reverting to Docker isn't the best move. At some point the container name changed from hometown_db_1 to hometown-db-1, for instance, leading to a moment of panic when I got a complaint that the container doesn't exist. In contrast, very few people would rename the db service in docker-compose.yml, so this new method is more likely to work.

This also implies an improved command to restore the database, though be warned that I have yet to test it:

sudo docker compose exec db pg_restore -U postgres --clean --if-exists -d postgres < ~/backup.dump