hyperstack-org/hyperstack

don't transmit full record on change - only send changed values

catmando opened this issue · 2 comments

We are currently sending full records down on every change. Why this is happening is not clear. Something changed probably during a policy update.

It seems that this fix is working (see notes below)

module ReactiveRecord
  class Broadcast

    def self.after_commit(operation, model)
      # Calling public_columns_hash once insures all policies are loaded
      # before the first broadcast.
      @public_columns_hash ||= ActiveRecord::Base.public_columns_hash
      Hyperstack::InternalPolicy.regulate_broadcast(model) do |data|
        # filter out any unchanged fields unless its a create operation in which case we want to send everything
        # allowed by the policies.
        filter_unchanged_fields!(model.class.primary_key.to_sym, data) if operation != :create
        puts "Broadcast aftercommit hook: #{data}" if Hyperstack::Connection.show_diagnostics

        if !Hyperstack.on_server? && Hyperstack::Connection.root_path
          send_to_server(operation, data, model.__synchromesh_update_time) rescue nil # fails if server no longer running so ignore
        else
          SendPacket.run(data, operation: operation, updated_at: model.__synchromesh_update_time)
        end
      end
    rescue ActiveRecord::StatementInvalid => e
      raise e unless e.message == "Could not find table 'hyperstack_connections'"
    end unless RUBY_ENGINE == 'opal'

    def self.filter_unchanged_fields!(primary_key, data)
      data[:record] = data[:record].filter do |field, _| 
        data[:previous_changes].key?(field) || field == primary_key
      end
    end unless RUBY_ENGINE == 'opal'
....

Adding this causes this spec to fail: ./spec/batch5/authorization_spec.rb:113

It seems that spec itself is very brittle, and should be rewritten. It fails in a couple of places because it is in fact assuming that all data is being sent down even fields that have not changed. The spec uses some methods to interrogate the data on the client, without using the normal client interface. I do not believe an actual working application would have any problems.

So the reason that this was changed to submit the entire record, is that when making multiple saves within an active-record transaction, only the last values saved by previous_changes will be saved. The ar_transaction_changes gem is a work around for this, but we should investigate a more efficient / complete solution.