kafka error when running in swarm mode

[BogdanIrimie]

When running the compose file example zookeper and kafka containers start ok, but when entering the kafka container and using ./kafka-console-producer.sh --broker-list <host_machine_ip>:9092 --topic test-kafka, the following error is generated:

WARN Error while fetching metadata with correlation id 1 : {test-kafka=LEADER_NOT_AVAILABLE} (org.apache.kafka.clients.NetworkClient)

Tried to change the host_mahcine_ip with conatiner_ip but still no success.

When running the consumer (./kafka-console-consumer.sh --zookeeper zookeeper:2181 --topic test-kafka --from-beginning) the following error appears

WARN [console-consumer-53007_b9e6b370721d-1502384702881-8827def7-leader-finder-thread]: Failed to add leader for partitions test-kafka-0; will retry (kafka.consumer.ConsumerFetcherManager$LeaderFinderThread)
kafka.common.UnknownTopicOrPartitionException

The objective is to run kafka in a cluster of 3 or more.

[nfazekas]

+1

[mrafeeq]

+1

[sscaling]

@IrimieBogdan - Did you try connecting to the public broker port 9094 ?

[ymatsiuk]

Had the same issue.
Check ./kafka-topics.sh --zookeeper zookeeper:2181 --describe
I assume you have no leader for partitions for particular topics.

Fixed that with

broker.id.generation.enable=false

and setting the following as broker ids:

export BROKER_ID=$(hostname -i | sed 's/\.//g')

(in my case this IP is bind to container, we're using host network to run kafka)
In compose file added as well:

      update_config:
        parallelism: 1
        delay: 5m
        order: stop-first

So next time container got killed/rescheduled or updated, first it stops broker gracefully and afterwards all partitions rebalanced.