The Crystal Programming Language Forum

Limit concurrency for HTTP::Client


I’m currently implementing a daily import jobs against several APIs using Crystal (no need to write Java and save some resources, right :slight_smile:)

While I would like to utilize concurrency and sending several requests at the same time, I do not want to overload the API endpoint and cap the number of concurrent requests. What would be a good approach to that?

The naive solution would be to have a list of items to process, shifting n elements of that list, send n fibers and use a channel to wait until all fibers have returned. That however means everything is dependent on the slowest fiber. Another idea would be to have block like

parallelize(10) do |e|
  # HTTP client request here

This would require some more synchronization. Another idea could be one of the rate limiter shards, but most of them are based on time (which makes sense for rate limiting i.e. with a token bucket).

I suspect I am thinking too complex and there is something easier to do, so happy to get better ideas.

Happy holidays to everyone!


1 Like

You are describing a fiber pool.
An implementation is

1 Like

Another implementation that I just put together

module Indexable(T)
  def fiberpool_map(workers : Int, &func : T -> R) forall T, R
    Array(R).build(self.size) do |result|
      ch = Channel({Int32, T}).new
      workers.times do
        spawn do
          while (work = ch.receive?)
            i, inp = work
            result[i] =
      self.each_with_index do |x, i|
        ch.send({i, x})
p (1..100).to_a.fiberpool_map(5, &.**(3))

(for actual parallelism see also: threadpool_map)


Hey Alex, I think you might find my rate_limiter ( useful to keep an eye on the number of API calls rate.

Going through the readme should be enough for you to figure out whether this is relevant to your use case :+1:

1 Like

Oops just realized that my implementation is broken because it doesn’t wait at the end.
Exercise for the reader ¯\_(ツ)_/¯ :thinking: