prototype wrapping of query execution for AR cursor

[eljojo]

hey @kirs @byroot i'm doing a prototype here, no real intentions of merging, any comments welcome. I am doing this for a personal project.

here's the context: I have a vanilla rails app and I'm trying to use multiple connections within the same job. I am using rails' connected_to block mechanism.

I am running into a problem where I switch connections during each_iteration, which makes the job finish after the first batch is done. It seems that AR switches the default to connection to whatever connected_to block was called last.

In any case, this is an attempt to make it possible to pass a block around AR enumerator builder, so I can wrap the query execution in the appropriate connected_to block.

usage is kinda weird but not the end of the world:

  def build_enumerator(klass_name, cursor:)
    enumerator_builder.active_record_on_records(
      klass_name.safe_constantize.all,
      cursor: cursor,
      around_query: -> (&block) { Shards.legacy(&block) }
    )
  end

[kirs]

Hmm. I'd love to make it work nicely with AR's multi-database feature. Ideally with developers not having to pass around_query:.

I think we could scope this out to vanilla ActiveRecord gist like:

enum = Enumerator.new do |yielder|
  ActiveRecord::Base.connected_to(role: :foo) do
     Shop.find_each { |shop| yielder.yield(record) }
  end
end

enum.each do
  ActiveRecord::Base.connected_to(role: :bar) do
    # do something else
  end
end

Would that work? If not maybe it's an edge case of ActiveRecord that we could take care of upstream.