6
votes

When trying to consume from Kafka using the high-level consumer (using a completely new consumer group), the consumer never starts running. When I switch the logging level to debug, I can see the following two lines repeating themselves over and over:

DEBUG [AbstractCoordinator] 09:43:51.192: Sending coordinator request for group CompletelyNewConsumerGroupThatIneverUsedBefore to broker 172.30.3.61:30000 (id: 1043 rack: null)
DEBUG [AbstractCoordinator] 09:43:51.306: Received group coordinator response ClientResponse(receivedTimeMs=1487666631306, disconnected=false, request=ClientRequest(expectResponse=true, callback=org.apache.kafka.clients.consumer.internals.ConsumerNetworkClient$RequestFutureCompletionHandler@58f68932, request=RequestSend(header={api_key=10,api_version=0,correlation_id=197,client_id=consumer-1}, body={group_id=CompletelyNewConsumerGroupThatIneverUsedBefore}), createdTimeMs=1487666631192, sendTimeMs=1487666631192), responseBody={error_code=15,coordinator={node_id=-1,host=,port=-1}})

All 6 nodes of the cluster consume data properly. Even the old low-level consumer works. Also in Zookeeper there are only the 6 nodes that should be there and their log files look ok.

From googling I found that the error_code=15 means that the Group coordinator is not available. I assume that means Consumer Group coordinator. Is this a callback problem? Why does the high-level consumer not work?

1
High-level consumer is the old consumer. Based on the exception thrown, you should be using the new consumer, and consumer complains it cannot find the coordinator. Did you use the auto-generated IDs for brokers?amethystic
@amethystic, hello, could you please point me to the doc where it says that old consumers complain about group coordinator? Is it valid for 0.9.x?Novitoll

1 Answers

6
votes

As it turned out, the all partitions of the __consumer_offsets topic were located on dead nodes (nodes that I turned off and that will never come back). I solved the issue by shutting the cluster down, deleting the __consumer_offsets topic from Zookeeper and then starting the cluster again.