add client_class option

[feymartynov]

A use case for it is setting up a connection to AWS Elasticsearch.

According to elasticsearch-transport's readme one should pass a block to ::ElasticsearchClient#initialize which adds necessary authorization headers to each request. However this can't be done with chewy.

A possible hack as it was discussed in the issue #296 is to assign Thread.current[:chewy_client] somewhere (say, in Rails initializer) with a custom instance. But this would affect only the current thread.

With :client_class option it becomes possible to perfom dependency injection of a custom client class which gets instantiated in every thread you call Chewy.client.

So an example initializer for AWS tuning looks like this:

require 'faraday_middleware/aws_signers_v4'

class AwsElasticsearchClient < ::Elasticsearch::Client
  def initialize(options)
    super(options) do |f|
      f.request :aws_signers_v4,
                service_name: 'es',
                region: 'us-east-1',
                credentials: Aws::Credentials.new(
                  ENV['AWS_ACCESS_KEY'],
                  ENV['AWS_SECRET_ACCESS_KEY'])
    end
  end
end

Chewy.configuration[:chewy_client] = AwsElasticsearchClient

[davekaro]

Awesome great idea. This would work perfect for #296

[pyromaniac]

Hm, I was mostly thinking about transport_config settings proc, which is passed on transport initializaton.

[feymartynov]

Created a new pull request for this: #302.

[morgler]

Thanks for the pull request! But I have trouble using it correctly :(. I do this in my chewy.rb initializer:

  Chewy.settings = {
    host: 'http://my-es-instance-on-aws.us-east-1.es.amazonaws.com'
  }
  Chewy.configuration[:transport_options] = {
    headers: { content_type: 'application/json' },
    proc: -> (f) do
        Rails.logger.debug ">>> Calling proc for chewy client"
        f.request :aws_signers_v4,
                  service_name: 'es',
                  region: 'us-east-1',
                  credentials: Aws::Credentials.new(
                    Rails.application.secrets.aws_access_key_id,
                    Rails.application.secrets.aws_secret_access_key)
    end
  }

If I try to use chewy afterwards, I always get a Elasticsearch::Transport::Transport::Errors::Forbidden error as if I would not have supplied the AWS credentials. I inserted a log line in the proc and it is never reached.

I also deleted chewy.yml to avoid collisions of it with the initializer.

What am I doing wrong? Maybe you could add an example to the readme of how to use this proc correctly.

[morgler]

Okay, rubber ducking helped ;). It seems you can't mix the configuration methods. So I had to write it all in one block:

  Chewy.settings = {
    host: 'http://my-es-instance-on-aws.us-east-1.es.amazonaws.com',
    transport_options: {
      headers: { content_type: 'application/json' },
      proc: -> (f) do
          Rails.logger.info ">>> Calling proc for chewy client"
          f.request :aws_signers_v4,
                    service_name: 'es',
                    region: 'us-east-1',
                    credentials: Aws::Credentials.new(
                      Rails.application.secrets.aws_access_key_id,
                      Rails.application.secrets.aws_secret_access_key)
      end
    }
  }