We recently upgraded to Rails 4.2.0.beta2 to take advantage of the new Active Job framework and other improvements. We chose Sucker Punch to avoid the cost of using a separate worker on Elastic Beanstalk, along with Fist Of Fury for scheduling ( which we ended up not using ).
"I use sucker_punch as a poor man's sidekiq. I can have async jobs on Heroku without paying the extra worker. My jobs are usually very small and fast, so it works just fine."
So we introduced active jobs to spawn threads for anything that takes time, like emails, external API calls, and heavy DB processing.
Having done all that, we were initially stumped as to why we saw no noticeable change in behaviour.
In the Rails log we saw entries like this, where the Active Job framework was en-queuing and immediately running the jobs in the same thread:
[ActiveJob] [MyJob] [ec05...7bd] Performed MyJob from Inline(default) in 949.05ms
The key word is “Inline” - This means the default queue adapter was being used - hence the jobs executed immediately. Our queue adapter configuration was located within an initialiser, but when we moved it to the main config/application.rb, it worked!
# config/application.rb module MyApp class Application < Rails::Application ... ... # config settings ... ... config.active_job.queue_adapter = :sucker_punch end end
The other thing we found is - you can pass active record objects as parameters to active jobs, so long as the data is persisted. If you try to pass an unsaved object, say if you are deferring saving active record objects using an active job, you will get an exception like this:
Expected a URI like gid://app/Person/1234: #<URI::Generic:0x85b2220 URL:gid://my-app/MyClass/>
The solution is to use some other data structure like a hash, or save the object before passing it as a parameter to the active job.