Adding support for parallel requests

!> This is slightly more involved, and this section is not fully formed.

Vague example, excerpted from the test suite about parallel requests

ruby

response_1 = nil
response_2 = nil

conn.in_parallel do
  response_1 = conn.get('/about')
  response_2 = conn.get('/products')
end

puts response_1.status
puts response_2.status

First, in your class definition, you can tell Faraday that your backend supports parallel operation:

ruby

class FlorpHttp < ::Faraday::Adapter
  dependency do
    require 'florp_http'
  end

  self.supports_parallel = true
end

Then, implement a method which returns a ParallelManager:

ruby

class FlorpHttp < ::Faraday::Adapter
  dependency do
    require 'florp_http'
  end

  self.supports_parallel = true

  def self.setup_parallel_manager(_options = nil)
    FlorpParallelManager.new # NB: we will need to define this
  end

  def call(env)
    # NB: you can call `in_parallel?` here to check if the current request
    # is part of a parallel batch. Useful if you need to collect all requests
    # into the ParallelManager before running them.
  end
end

class FlorpParallelManager
  # The execute method will be passed the same block as `in_parallel`,
  # so you can either collect the requests or just wrap them into a wrapper,
  # depending on how your adapter works.
  def execute(&block)
    run_async(&block)
  end
end

A note on the old, deprecated interface

Prior to the introduction of the execute method, the ParallelManager was expected to implement a run method and the execution of the block was done by the Faraday connection BEFORE calling that method.

This approach made the ParallelManager implementation harder and forced you to keep state around. The new execute implementation allows to avoid this shortfall and support different flows.

As of Faraday 2.0, run is still supported in case execute is not implemented by the ParallelManager, but this method should be considered deprecated.

For reference, please see an example using run from em-synchrony and its ParallelManager implementation.