Kafka consumer fetching metadata for topics failed

Question

I am attempting to write a Java client for a third party's Kafka and ZooKeeper servers. I am able to list and describe topics, but when I attempt to read any, a ClosedChannelException is raised. I reproduce them here with the command line client.

$ bin/kafka-console-consumer.sh --zookeeper 255.255.255.255:2181 --topic eventbustopic
[2015-06-02 16:23:04,375] WARN Fetching topic metadata with correlation id 0 for topics [Set(eventbustopic)] from broker [id:1,host:SOME_HOST,port:9092] failed (kafka.client.ClientUtils$)
java.nio.channels.ClosedChannelException                                       
    at kafka.network.BlockingChannel.send(BlockingChannel.scala:100)           
    at kafka.producer.SyncProducer.liftedTree1$1(SyncProducer.scala:73)        
    at kafka.producer.SyncProducer.kafka$producer$SyncProducer$$doSend(SyncProducer.scala:72)
    at kafka.producer.SyncProducer.send(SyncProducer.scala:113)                
    at kafka.client.ClientUtils$.fetchTopicMetadata(ClientUtils.scala:58)        
    at kafka.client.ClientUtils$.fetchTopicMetadata(ClientUtils.scala:93)        
    at kafka.consumer.ConsumerFetcherManager$LeaderFinderThread.doWork(ConsumerFetcherManager.scala:66)
    at kafka.utils.ShutdownableThread.run(ShutdownableThread.scala:60)         
[2015-06-02 16:23:04,515] WARN Fetching topic metadata with correlation id 0 for topics [Set(eventbustopic)] from broker [id:0,host:SOME_HOST,port:9092] failed (kafka.client.ClientUtils$)
java.nio.channels.ClosedChannelException                                       
    at kafka.network.BlockingChannel.send(BlockingChannel.scala:100)           
    at kafka.producer.SyncProducer.liftedTree1$1(SyncProducer.scala:73)        
    at kafka.producer.SyncProducer.kafka$producer$SyncProducer$$doSend(SyncProducer.scala:72)
    at kafka.producer.SyncProducer.send(SyncProducer.scala:113)                
    at kafka.client.ClientUtils$.fetchTopicMetadata(ClientUtils.scala:58)        
    at kafka.client.ClientUtils$.fetchTopicMetadata(ClientUtils.scala:93)        
    at kafka.consumer.ConsumerFetcherManager$LeaderFinderThread.doWork(ConsumerFetcherManager.scala:66)
    at kafka.utils.ShutdownableThread.run(ShutdownableThread.scala:60)         

Alternate commands succeed:

$ bin/kafka-topics.sh --describe --zookeeper 255.255.255.255:2181 --topic eventbustopic
Topic:eventbustopic   PartitionCount:2        ReplicationFactor:1     Configs:
    Topic: eventbustopic  Partition: 0    Leader: 1       Replicas: 1     Isr: 1
    Topic: eventbustopic  Partition: 1    Leader: 0       Replicas: 0     Isr: 0

$ bin/kafka-topics.sh --list --zookeeper 255.255.255.255:2181 --topic eventbustopic
eventbustopic

(The ips were redacted and replaced with 255.255.255.255)

When I google this exception, I see issues on the producer side -- indeed, the source for ClientUtils.fetchTopicMetadata suggests this is mainly used by producers.

One concern that I have is that this might be a product of the network layout: the packets are mangled by Haproxy and sent over a VPN.

What exactly is at work here?

Answer 1

The broker tells the client which hostname should be used to produce/consume messages. By default Kafka uses the hostname of the system it runs on. If this hostname can not be resolved by the client side you get this exception.

You can try setting advertised.host.name in the Kafka configuration to an hostname/address which the clients should use.