Во время выполнения кода

TableSync::Publishing::Single
  .new(object_class: 'Player::FraudStatus', original_attributes: data, debounce_time: 0)
  .publish_later

Возникает ошибка ArgumentError: Cache expiration time is invalid, cannot be negative: -0.000102
Судя по всему ошибка возникает тут - https://github.com/umbrellio/table_sync/blob/master/lib/table_sync/publishing/helpers/debounce.rb#L130 , потому что ActiveSupport cache не может переварить Time.current .

discuss the idea of renaming table_sync -> rofl_sync

in #36 I created docs/message_protocol.md but it is not enough

subj

It is suitable for "temporary" ignore some fields in order to make some technical works with some fields during the regular working sync) for example: email delivery over the big amount of clients with temporary changed email field during the active updates of the client table (but we want the email field to remain non-changeable)

At the moment, to stimulate this, we need to build complex combinations of wrap_receiving, additional_data, mapping_overrides, only and etc.

It would be nice to have the new option with both dynamic and static behavior (in order to calculate ingorance list in runtime or in load time)

todo

We have only:

class Rabbit::Handler::MyProject::TableSync < TableSync::ReceivingHandler
   receive "User", to_table: :users do
      only %w[project_id user_name]
   end
end

It would be convenient to have except too:

class Rabbit::Handler::MyProject::TableSync < TableSync::ReceivingHandler
   receive "User", to_table: :users do
      except %w[user_age]
   end
end

TableSync defines hooks directly in model without including or prepending code and if you define similar hooks, tablesync's hooks are shadowed and don't work

class Test < Sequel::Model
  TableSync.sync(self)

  def after_create
    super
    p "AFTER CREATE"
  end
end

Test.create

You will see "AFTER CREATE" in console but data won't be synced

subj

1. Не понимаю, что проверяют спеки avoid dead locks, так как там переопределяется upsert и мы там не берем вообще никаких локов. К тому же эти спеки отжирают больше 3/4 времени выполнения вообще всех спеков из-за sleep 2.
2. В Актив Рекорде берется куча локов под транзой FOR NO KEY UPDATE и как раз там вероятность дедлока крайне высока?
3. В Сиквеле не берется никаких локов, соотвественно, победит та транза, которая коммитнется последней и перетрет изменения предыдущей транзы
4. Кажется, вместо кучи локов в актив рекорде и отсутсвии этих локов в сиквеле имеет смысл изменить уровень изоляции транзы на какую-нибудь сериализацию 🤔

Potential Concept:

  receive_no_sync "User", do |data, receiving|
     # your code here
  end

(at this moment we can realize this logic only via wrap_receiving instruction)

Current implementation intends that update and destroy events must have same attribute fields.
As consequence, dirty additional_data appears.
I think it will be useful if it is possible to create additional_data for each type of event separately.

Below case that shows dirty additional_data.
Let's imagine that receiver get update-event with attributes:

   {
     id: 12,
     name: "kek"
     ...,
   }

and destroy_event with:

   {
     id: 12,
   }

receiver is:

receive "Model", to_table: :models do
  additional_data do |current_row:|
    next {} if current_row.keys.count == 1 # destroy condition
    super_name = current_row[:name].upcase
    {
      name: super_name
    }
  end
  on_destroy do |attributes:, target_keys:| # rubocop:disable Lint/UnusedBlockArgument
    Model[attributes[:id]].update
  end
end

TableSync::Plugins it is not plugins, It is API for monkey patching which makes monkey patching difficult and does not resolve problems with monkey patching

absolutely useless

Now:
If I setup synchronization with if predicate like
TableSync.sync(self, if: -> (*) { syncable? })
model will send updates if predicate is truthy, but destroy will always send.

What I want:
The model must check predicates when sending both types of updates.

1. add message versions and a counter that will send errors for message omissions
2. add feature that allows to send more than one model in batches. it resolve some consistency issues.
3. add information to readme, why should we use batches and what problems does it solve.

at now destroy event is ignoring version_key but it is wrong behavior for obvious reasons

We want to have an ability to define/configure an individual logger for each receiver separately (incoming messages, malformed messages, etc). this is a good way to disable logging for unnecessary (or very active) receivers on staging systems (you can disable some super-active logging operations, but not the hole logging system).

1. Maybe we should rename it to something like routing_headers_callable? Right now it's not very clear that it only affects RabbitMQ headers.
2. Currently it yields only model class name and raw attributes, but I might have attributes_for_sync method defined so I need to do stuff like model = model_class.constantize.new(attrs).try(:attributes_for_sync) which is slow and inefficient. Would be cool to be able to get processed attributes there as well (or instead of raw attributes).

Monotonic Clock is not affected by discontinuous jumps in the system time e.g., if the system administrator manually changes the clock