Migrating from Sidekiq to Solid Queue with Scheduling
This is going to be a more technical post than what you’re used to reading here, but it’s quite a big change, so I believe it’s the right place to share it.
Solid Queue
I, like many other Rails developers, was using Sidekiq for background jobs. It’s a great gem, but unless you’re paying for the pro version, it’s quite unsafe. When Sidekiq starts processing a job it removes it from Redis. If the worker is interrupted and/or killed before the job is finished, that job will be lost. The solution is upgrading to the pro version, but that is quite expensive for a small project like Visualizer.
So when David introduced Solid Queue at Rails World 2023, it looked like a perfect solution. It was lacking a few crucial things, though. The first one being a simple web interface to see the jobs. That got quickly resolved with Mission Control - Jobs. It’s very simple, but it gets the job done. No pun intended.
The other big thing was scheduling jobs. I was using sidekiq-scheduler for this before, and it’s a very important feature to have. So I was closely following the PR#155 and yesterday Rosa merged it in. 🥳
Literally minutes later Visualizer was running it in production, and it’s been working great since.
The Migration
At first, I had to point directly to the GitHub repository, but the updates have now been released, so all you need is to replace sidekiq* entries in your Gemfile
with:
gem "solid_queue", ">= 0.3.0"
gem "mission_control-jobs", ">= 0.2.0"
Then you need to run bundle install
, bin/rails generate solid_queue:install
, and bin/rails db:migrate
, to prepare the database and generate the configuration file.
After that you should set the queue adapter in config/environments/production.rb
:
config.active_job.queue_adapter = :solid_queue
Finally, you need to expose jobs in your config/routes.rb
. Ideally behind some kind of authentication:
authenticate :user, ->(user) { user.admin? } do
mount MissionControl::Jobs::Engine, at: "/jobs"
end
At this point, you can find and delete the rest of sidekick related code in your project, and you should be good to go.
Configuring Recurring Jobs
The scheduling is done in config/solid_queue.yml
and it’s quite flexible. The schedule
key can take anything Fugit::Cron
can parse. The class
is the job you want to run and the optional args
are the arguments you can pass to the job. It can even take kwargs as the last parameter. That all goes under the recurring_tasks
inside dispatchers
. Here’s an example from Visualizer:
default: &default
dispatchers:
- polling_interval: 1
recurring_tasks:
shared_shot_cleanup_job:
class: SharedShotCleanupJob
schedule: "@hourly"
duplicate_stripe_subscriptions_job:
class: DuplicateStripeSubscriptionsJob
schedule: "@daily"
airtable_webhook_refresh_all_job:
class: AirtableWebhookRefreshAllJob
schedule: "0 0 */6 * *"
Error Handling
Because Solid Queue is a new gem, you can’t assume your error tracking service will support it out of the box. For example, I’m using AppSignal, and they don’t support it yet. There’s an open issue, but not much progress yet. At least not publicly.
Luckily this gem is build by Rails core team, so they provide a very simple on_thread_error
configuration option. In config/environments/production.rb
, right under the queue adapter configuration, you can add:
config.solid_queue.on_thread_error = ->(error) { Appsignal.send_error(error) }
And for errors that happen in jobs, you can hook in to ApplicationJob
and ApplicationMailer
adding this snippet:
rescue_from(Exception) do |exception|
Appsignal.send_error(exception)
raise exception
end
This will send all Solid Queue and job errors to AppSignal, but you can easily change it to your own error tracking service.
Conclusion
It has not even been a day, but I’m already very happy with the switch. If nothing else it decreases my dependence on Redis. Not that I have anything against Redis, but it’s one more thing to maintain and keep an eye on.
Now the last piece of the puzzle will be switching to Solid Cache, and then I can finally say goodbye to Redis. Less service dependencies, fewer things to worry about, better uptime, happier users. 🚀
In Other News
As usual, there were many other changes since the last update, with the most significant being the introduction of a daily cap of 50 shots on the free plan. This adjustment is expected to affect only a very small fraction of users. However, those who exceed this limit are encouraged to upgrade to the premium plan, as their usage places considerable strain on the database.
Thank you! 🙏