Should rd_kafka_flush always call poll?

[coldeasy]

Description

rd_kafka_flush does not call rd_kafka_poll if the timeout given is zero. I feel like it should always call it once since if someone is trying to use this in a non-blocking way, repeatedly calling rd_kafka_flush with a zero timeout will not give the desired results. We can easily work around this by calling rd_kafka_poll repeatedly and checking the queue length afterwards however I think making rd_kafka_flush call rd_kafka_poll at least once would yield a nicer API to work with.

How to reproduce

Produce some messages
Repeatedly call rd_kafka_flush
Notice that the queue length does not decrease.

Checklist

Please provide the following information:

• librdkafka version (release number or git tag): v0.10.5
• [x ] Apache Kafka version: 0.10.2
• librdkafka client configuration: <REPLACE with e.g., message.timeout.ms=123, auto.reset.offset=earliest, ..>
• [x ] Operating system: Centos 6 (x64)
• Provide logs (with debug=.. as necessary) from librdkafka
• Provide broker log excerpts
• Critical issue

[edenhill]

That makes sense, will fix 👍

[coldeasy]

👍 Thanks!

[edenhill]

Could you try out the fix (on the fetcholdrktp branch)?

[coldeasy]

I've tested it with the python client and it works well. Thanks for the quick turnaround.

Before

>>> producer = confluent_kafka.Producer({'bootstrap.servers': '{}:{}'.format(kafka_host, kafka_port)})
>>> producer.produce('topic', 'data')
>>> producer.produce('topic', 'data')
>>> len(producer)
3
>>> producer.flush(0)
3L
>>> len(producer)
3

After

>>> producer = confluent_kafka.Producer({'bootstrap.servers': '{}:{}'.format(kafka_host, kafka_port)})
>>> producer.produce('topic', 'data')
>>> producer.produce('topic', 'data')
>>> len(producer)
4
>>> producer.flush(0)
0
>>> len(producer)
0

[edenhill]

Thank you!