Troubleshooting errors

IPv4/IPv6 conflicts

Having IPv6 on your DNS, but not getting the correct IP or firewalling applied can cause problems including performance (timing out errors) or failing jobs for the fediverse. Make sure you have fully working dual stack if you have a ‘AAAA’ record in your DNS zone.

I see an error page that says something went wrong. How do I find out what’s wrong?

All error messages with stack traces are written to the system log. When using systemd, the logs of each systemd service can be browsed with journalctl -u mastodon-web (substitute with the correct service name). When using Docker, it’s similar: docker logs mastodon_web_1 (substitute with the correct container name).

Specific details of server-side errors are never displayed to the public, as they can reveal what your setup looks like internally and give attackers clues on how to get in, or how to abuse the system more efficiently.

Each response from Mastodon’s web server carries a header with a unique request ID, which is also reflected in the logs. By inspecting the headers of the error page, you can easily find the corresponding stack trace in the log.

I’m not seeing much in my logs. How do I enable additional logging/debugging information?

Please note that Mastodon (as a 12-factor application), logs to stdout, and NOT #{Rails.root}/log/#{Rails.env}.log as it’s usual in Rails. Watch your console!

By default your logs will show info level logging. To see more debugging messages, you can your .env.production file to increase the level, for the relevant service:

  • Web/Sidekiq: Set the value of RAILS_LOG_LEVEL to debug and then restart the service that you’re attempting to troubleshoot.
  • Streaming: Set the value of LOG_LEVEL to silly and then restart the service that you’re attempting to troubleshoot.

More information on other logging levels for these option can be found on the Configuring your environment page.

The debug or silly levels can be very verbose and you should take care to change the log level back to a lower level, once you have completed your troubleshooting.

After an upgrade to a newer version, some pages look weird, like they have unstyled elements. Why?

Check that you have run RAILS_ENV=production bin/rails assets:precompile after the upgrade, and restarted Mastodon’s web process, because it looks like it’s serving outdated stylesheets and scripts. It’s also possible that the precompilation fails due to a lack of RAM, as webpack is unfortunately extremely memory-hungry. If that is the case, make sure you have some swap space assigned. Alternatively, it’s possible to precompile the assets on a different machine, then copy over the public/packs directory.

After an upgrade to a newer version, some requests fail and the logs show error messages about missing columns or tables. Why?

Check that you have run RAILS_ENV=production bin/rails db:migrate after the upgrade, because it looks like Mastodon’s code is accessing a newer or older database schema. If you are using PgBouncer, make sure this one command connects directly to PostgreSQL, as PgBouncer does not support the kind of table locks that are used within migrations.

I am trying to run a tootctl or rake/rails command, but all I get is an error about uninitialized constants. What’s wrong?

Check that you are specifying the correct environment with RAILS_ENV=production before the command. By default, the environment is assumed to be development, so the code tries to load development-related gems. However, in production environments, we avoid installing those gems, and that’s where the error comes from.

I encountered a compilation error while executing RAILS_ENV=production bundle exec rails assets:precompile, but no more information is given. How to fix it?

Usually, it’s because your server ran out of memory while compiling assets. Use a swapfile or increase the swap space to increase the memory capacity. Run RAILS_ENV=production bundle exec rake tmp:cache:clear to clear cache, then execute RAILS_ENV=production bundle exec rails assets:precompile to compile again. Make sure you clear the cache after a compilation error, or it will show “Everything’s OK” but leave the assets unchanged.

I am getting this error: Read-only file system @ dir_s_mkdir. Why?

By default, Mastodon makes use of systemd’s sandboxing capabilities in a way that disallows writing outside of /home/mastodon. If Mastodon is installed elsewhere, you may need to allow mastodon-sidekiq and mastodon-web to write to a custom directory:

  1. Add parameter ReadWritePaths to files /etc/systemd/system/mastodon-sidekiq.service and /etc/systemd/system/mastodon-web.service. Example - ReadWritePaths=/example/mastodon/live.
  2. run systemctl stop mastodon-sidekiq mastodon-web
  3. run systemctl daemon-reload
  4. run systemctl start mastodon-sidekiq mastodon-web

The Mastodon WebUI just displays an error page, but the API works

This happens for example, when you run

RAILS_ENV=production bundle exec rails assets:precompile

without a valid, functional NodeJS environment. In some cases yarn install --immutable has not been run, yet and the web files have not been installed to the correct directory, yet. Running the assets::precompile command again will probably not help. Here’s how to fix it:

cd live
yarn install --immutable
RAILS_ENV=production bundle exec rake tmp:cache:clear
RAILS_ENV=production bundle exec rails assets:precompile

Then restart the mastodon-web service:

sudo systemctl restart mastodon-web.service

Postgres database migrations fail during a Mastodon update

If you encounter error messages like this:

WARNING:  you don't own a lock of type ExclusiveLock                                           
bin/rails aborted!                                                                             
ActiveRecord::ConcurrentMigrationError:  (ActiveRecord::ConcurrentMigrationError)              
                                                                                                                                                                                               
Failed to release advisory lock

You’re probably running pg_bouncer in front of PostgreSQL and forgot to specify the PostgreSQL port manually (instead of using the default pg_bouncer port!). To make the migrate commands work, add DB_PORT, for example:

RAILS_ENV=production DB_PORT=5432 bundle exec rails db:migrate

This needs to be done for both - pre-deployment migrations and post-deployment migrations. Also see: Scaling Mastodon - pg_bouncer section.

Last updated