Site automatically exits read-only mode when enabled by 'discourse enable_readonly'

[RhysGoodwin]

Hi folks,

I’m setting up some automation for deployment and migration with Ansible but I’ve got an issue with read-only mode.

When I enabled read-only with:
docker exec app discourse enable_readonly
or

./launcher enter app
discourse enable_readonly

After ~1:20 the site exits read-only mode all by itself.

If I enable read-only mode via the WebUI (admin/backups section), it stays enabled until I disable it.

Any thoughts?

[Lilly (Lillian)]

Did you check your logs? perhaps you have a background job or plugin somewhere that is interrupting the process.

[RhysGoodwin]

Hi Lilly,

Thanks for your response. This is a fresh default install and I’ve not installed any plugins or created any jobs. Which logs am I best to look at? Sorry I’m a bit of a noob.

I’ve tried tail -f on:

/var/discourse/shared/standalone/log/rails/unicorn.stdout.log
/var/discourse/shared/standalone/log/rails/unicorn.stderr.log
/var/discourse/shared/standalone/log/rails/production_errors.log
/var/discourse/shared/standalone/log/rails/production.log
/var/discourse/shared/standalone/log/rails/sidekiq.log
/var/discourse/shared/standalone/log/var-log/redis/current

I waited for the event to occur while following the logs but I didn’t see anything that stood out.

It’s odd to me that it works perfectly when activated via the WebUI but not via cli.

[pfaffman (Jay Pfaffman)]

I have had trouble using that as well, both for enabling and disabling readonly mode. The last time I ended up running Discourse.enable_readonly_mode in rails.

It’s always an emergency when it happens and I haven’t investigated what the issue might be.

FWIW, in my own Ansible tooling I don’t use read-only mode, but do use Introducing Post Deployment Migration

[RhysGoodwin]

Thanks Jay. I suspect I’m taking an XY approach to the problem. I’ll check out the Post Deployment Migration. Thanks for that.

I guess the other option might be to activate RO via an API call.

[pfaffman (Jay Pfaffman)]

- name: post migrations
  shell: "docker exec  {{ discourse_yml }} bash -c 'rake db:ensure_post_migrations db:migrate'"
  become: yes

so this might work for read-only mode:

- name: post migrations
  shell: "docker exec  {{ discourse_yml }} bash -c 'echo Discourse.enable_readonly_mode | rails c"
  become: yes

[RhysGoodwin]

Thank you, much appreciated. I’ll give this a go tonight.

[JammyDodger]

Is the difference down to the read-only mode you’re using?

@SaraDev

• READONLY_MODE_KEY is mainly used for the backup and restore process and is triggered by the application itself. This mode can also be enabled or disabled from the the Discourse command line interface with discourse enable_readonly and discourse disable_readonly. However, this key will not survive a container restart.
• USER_READONLY_MODE_KEY is used when an admin clicks the readonly only button in the admin interface. The special thing about this key is that we do not set it as an expiring key since readonly enabled by a user needs to survive container restarts. Other keys are set with a TTL of 60 seconds and we have a thread to extend the expiry every 30 seconds to ensure that an app is never stuck in readonly mode.

[RhysGoodwin]

Thanks JammyDodger! The info and link were very helpful. I’m not sure what constitutes a container restart - as far as I know my container is not restarting every ~1:30 which is as long as the RO mode stays enabled when using discourse enable_readonly i.e. Discourse.enable_readonly_mode(Discourse::READONLY_MODE_KEY)

I can confirm however that Discourse.enable_readonly_mode(Discourse::USER_READONLY_MODE_KEY) “sticks” indefinitely, which is what I was trying to achieve

[RhysGoodwin]

For completeness:

- name: Set read-only mode for live forum host
  ansible.builtin.shell: |
      docker exec app bash -c 'echo Discourse.enable_readonly_mode\(Discourse::USER_READONLY_MODE_KEY\) | rails c'
  register: enable_readonly_output