High volume metadata requests

[mjquinlan]

Hi Hernán,

In one of my applications, I've implemented a procedure of calling kafkerl:request_metadata() whenever I receive a Kafka based error to a data request. I've noticed that when the volume of requests are high during a cluster rebalance, the metadata request isn't being made. Instead the cooldown logic is blocked in this case. I suggest the following:

https://github.com/HernanRivasAcosta/kafkerl/blob/master/src/kafkerl_connector.erl#L341

case Cooldown - (Now - LastRequest) of
  Negative when Negative =< 0 ->
    _ = make_metadata_request(State),
    State#state{broker_mapping = void, known_topics = NewKnownTopics,
              last_metadata_request = Now};
  Time ->
    _ = timer:apply_after(Time, ?MODULE, request_metadata, [self(), true])
    State#state{broker_mapping = void, known_topics = NewKnownTopics,
              last_metadata_request = LastRequest}
end.