[librdkafka issue] rd_kafka_consumer_close() would keep retrying the "coordinator query" if there's no response from brokers -- thus the consumer would stuck there forever

[kenneth-jia]

• A placeholder
  librdkafka issue: rd_kafka_consumer_close() would keep retrying the "coordinator query" if there's no response from brokers -- thus the consumer would stuck there forever confluentinc/librdkafka#3740

• Description
  We found the PR OffsetCommit retry fixes confluentinc/librdkafka#3416 has changed the retry behavior for "offset commit".
  If a consumer try to commit offsets with rd_kafka_commit_queue(...) before closing (with rd_kafka_consumer_close(...)), but there's no response from brokers, the consumer would be keep retrying the "coordinator query" forever, thus the consumer would stuck there (and user's application could never quit)!

• How to reproduce

  • Robustness test: TEST(KafkaConsumer, DISABLED_AlwaysFinishClosing_ManuallyPollEvents)
  • Robustness test: TEST(KafkaConsumer, DISABLED_CommitOffsetWhileBrokersStop)