Users joa1 and markb1 report inconsistent site settings behavior on Ubuntu development installations of Discourse, where changes are not reflected after refresh — likely due to caching or worker isolation. markb1 observes that reducing Puma workers to 1 resolves the issue locally but not in scaled environments, suggesting each worker holds its own settings cache. Falco clarifies that the official installation method doesn’t use Puma and doesn’t exhibit this bug, implying the issue stems from dev setup. markb1 confirms using the Ubuntu dev guide and launching via bundle exec rails server, which defaults to Puma. pfaffman suggests verifying server status and testing settings via Rails console. riking identifies a broken MessageBus — likely due to misconfigured Redis — as the root cause, noting live updates may also fail. markb1 adds that site_setting_saved events are still triggered but no longer listened to, possibly indicating outdated code paths. Falco suggests using Unicorn (bin/unicorn) as an alternative in dev. The core issue appears tied to Puma + Redis + MessageBus in non-standard dev setups, not production installations.
I’m seeing the same thing. If I set the Puma worker count to 1 the issue goes away. It appears that each worker is holding a copy of the site settings object, and when one worker performs an update the other workers are never aware of that update.
Unfortunately, this issue is also present in a scaled-out environment with multiple Discourse servers, so just setting the worker count to 1 isn’t a total fix.
I’m trying to do some plugin development, so I followed the link in the README to setup an Ubuntu development environment. When I run the command bundle exec rails server --binding=0.0.0.0 it starts the app with Puma.
The server is still running. I’ve verified from the logs that Discourse is serving the request. I also have cache disabled in my browser’s developer console while I’m debugging this.
Why are you scaling out during plugin development?
The symptoms indicate that your internal MessageBus is broken, so something’s wrong with your Redis setup. Live updates in the browser may also be broken.
I wasn’t scaling out for plugin development, just testing this in a few different scenarios to try to narrow down my issue and make sure I understood it. If the MessageBus is supposed to handle this then I will look into that further.
While debugging, I did notice that app/models/site_setting.rb is still triggering a :site_setting_saved event, which appears to have been removed from the app and replaced with :site_setting_changed. There is nothing listening to the :site_setting_saved event anymore.