Fixes segfault on SIGINT

[attdona]

• Removed all finalizer call because for what I know it is hard to say in what orders finalizer get called
  and this may cause problems with the correct order of api call expected by librdkafka.

• added kafka_consumer_close

• removed timer kafka_poll from KafkaClient

[dfdx]

Could you please describe what kind of issues you got with the kafka_poll timer? I may have misused the Timer type, but I think we still need to call kafka_poll at regular intervals to make callbacks and other stuff work correctly (e.g. see #13).

[attdona]

If you call rd_kafka_destroy, both via a finalizer or with an explicit close method, than an eventual timer callback will cause a segfault: the call to rd_kafka_destroy destructor has terminated the kafka threads when the timer triggers the callback.

I think this is the reason why a segfault is not always reproducible: it depends if the timer triggers the callback after client termination.

I don't see any reason to call a poll in a background task, are you sure it is needed?

[dfdx]

According to rdkafka.h (which is one of the primary sources of documentation btw):

@remark An application should make sure to call poll() at regular
intervals to serve any queued callbacks waiting to be called.
@remark If your producer doesn't have any callback set (in particular
via rd_kafka_conf_set_dr_msg_cb or rd_kafka_conf_set_error_cb)
you might chose not to call poll(), though this is not
recommended.

Similar to what's proposed in #13, I think we can add a flag .polling to the KafkaClient and use it to stop polling just before calling rd_kafka_destroy() in Base.close(::KafkaClient). Do you think it will be sufficient to avoid the conflicts?

[kschzt]

Indeed calling rd_kafka_poll() is required for delivery reports and error callbacks. If it isn't called, the application won't receive errors from librdkafka and can end up in a dead state. As a Producer, there are no alternatives to rd_kafka_poll(), as a Consumer you can get away with only rd_kafka_consumer_poll() if you call rd_kafka_poll_set_consumer() to redirect the callbacks to it. I've chosen to just call rd_kafka_poll() in both cases in #13. HTH

[attdona]

Nice, in #13 rd_kafka_poll is protected and it will no called after kafka destructor, but it is called inside a thread: If SIGINT is delivered to that thread this prevents a clean shutdown of the application.

Should it work if InterruptException is catch and rethrow?

Also a Timer may cause problems with SIGINT, see here.

[dfdx]

I've just merged #13 with polling in the @async-created task (i.e. in the same thread). If I'm not missing anything, rebasing this PR on master now should be sufficient to to avoid segfault. @attdona Do you see any other issues?

[attdona]

Thanks! Now the segfault issue is resolved. I've posted a note in #13 regardind the use of an @async task inside the library that is a possible source of the fatal: error thrown and no exception handler available uncatchable exception.