Enqueue Jobs Quickly with Sidekiq’s Bulk Features

Our focus is often on the potentially large queues or the time it will take to churn through all the work. But there’s another part of this that we have to consider: enqueuing all that work in the first place.

Instead of…

…looping through lots of Active Record objects:

Customer.all.each do |customer|
  DeliverEmailJob.perform_async(customer)
end

…or better, using find_each, if you have more than a few hundred jobs to enqueue:

Customer.find_each do |customer|
  DeliverEmailJob.perform_async(customer)
end

Use…

…Sidekiq directly:

Customer.all.in_batches do |relation|
  array_of_args = relation.ids.map { |x| [x] }
  DeliverEmailJob.perform_bulk(array_of_args)
end

Or, if using a Sidekiq version before 6.3.0:

Customer.all.in_batches do |relation|
  array_of_args = relation.ids.map { |x| [x] }
  Sidekiq::Client.push_bulk('class' => DeliverEmailJob, 'args' => array_of_args)
end

Why?

This minimises round trips to Redis. Instead of making an individual call for each enqueing action across the network, you make one. This is a big deal even when using tools as well optimised for speed as Redis.

This can also have the benefit of minimising the amount of memory used. By plucking ids, or using the ids method as shown, rather than looping over arrays of Active Record models, you are using less complex Ruby objects and thus less memory.

The author of Sidekiq recommends a limit of 1,000 jobs per bulk enqueue and that’s the default in the perform_bulk method. Even then you are saving the time of 999 round trips to Redis.

Why not?

You might, very sensibly, never enqueue huge numbers of jobs during a web request from a user, so why do we need to speed this up? Your customers are not necessarily suffering through the slower approach. However you are probably enqueuing jobs on the command line or in scheduled tasks so it’s now your time you are wasting!

The pre-6.3.0 version of the bulk API is a little fiddly and leaves space for you to make a mess of the method. I’d recommend updating to the latest version and then using the .perform_bulk syntax.

Also, this functionality requires that you use Sidekiq directly and not via Active Job. I have previously suggested plenty of reasons to use Sidekiq directly, primarily improved speed and flexibility.

Recommendation

If you’re using Sidekiq in production you should buy a license for the Pro version. The extra reliability so that jobs are not “lost” due to crashes is enough of a reason, but there are other additional features including job deletion (by job id and kind of job) and web dashboard enhancements.

Disclamer

I’m a massive fan and “Pro” customer of Sidekiq and have met Mike (its author) at a couple of Ruby conferences. That hasn’t coloured my view of the library. Sidekiq has always been terrific and worth every penny for how we use it at work.