Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

Why could Kafka warn "partitions have leader brokers without a matching listener"?

I'm trying to get Kafka to work on docker-compose for the first time. The application runs fine without docker. But on docker, I get the error as described below. Any reason why Kafka would throw this error?

The error:

email-service_1 | 2018-12-01 14:32:02.448 WARN 1 --- [ntainer#0-0-C-1] o.a.k.c.NetworkClient : [Consumer clientId=consumer-2, groupId=kafka] 1 partitions have leader brokers without a matching listener, including [email-token-0]

My docker-compose config:

version: '3.3'
    services:
     zookeeper:
      image: wurstmeister/zookeeper
      ports:
       - "2181:2181"

 kafka:
  image: wurstmeister/kafka
  command: [start-kafka.sh]
  environment:
   KAFKA_ZOOKEEPER_CONNECT: zookeeper:2181
   KAFKA_ADVERTISED_HOST_NAME: 192.168.23.134
   KAFKA_CREATE_TOPICS: "email-token:1:1"
  volumes:
   - /var/run/docker.sock:/var/run/docker.sock
  ports:
   - "9092:9092"
  depends_on:
   - zookeeper

 email-service:
  build: ./email-service
  environment:
   SPRING_KAFKA_BOOTSTRAPSERVERS: kafka:9092
  ports:
   - "8081:8081"
  depends_on:
   - kafka
like image 495
user2997204 Avatar asked Dec 01 '18 14:12

user2997204


People also ask

How many brokers will be marked as leaders for a partition in Kafka?

Each partition has one server which acts as the "leader" and zero or more servers which act as "followers". The leader handles all read and write requests for the partition while the followers passively replicate the leader. If the leader fails, one of the followers will automatically become the new leader.

Why are there brokers in Kafka?

A Kafka broker allows consumers to fetch messages by topic, partition and offset. Kafka brokers can create a Kafka cluster by sharing information between each other directly or indirectly using Zookeeper. A Kafka cluster has exactly one broker that acts as the Controller.

What is Kafka partition leader?

Kafka - (Partition|Write) LeaderA leader handles all read and write requests for a partition while the followers passively replicate the leader. Each server acts as a leader for some of its partitions and a follower for others so load is well balanced within the cluster.

What happens when broker is down in Kafka?

During a broker outage, all partition replicas on the broker become unavailable, so the affected partitions' availability is determined by the existence and status of their other replicas. If a partition has no additional replicas, the partition becomes unavailable.


3 Answers

As stated in the comments to your question the problem seems to be with the advertised name for the Kafka broker. According to your docker-compose you should be using 192.168.23.134 but your email-service is using kafka:9092. You can try with this docker-compose. I replaced the wurstmeister services with the latest Zookeeper and Kafka provided by confluentinc and added your email-service.

---
version: '2'
services:
zookeeper:
    image: confluentinc/cp-zookeeper:latest
    environment:
    ZOOKEEPER_CLIENT_PORT: 2181
    ZOOKEEPER_TICK_TIME: 2000

kafka:
    image: confluentinc/cp-kafka:latest
    depends_on:
    - zookeeper
    ports:
    - 9092:9092
    environment:
    KAFKA_BROKER_ID: 1
    KAFKA_ZOOKEEPER_CONNECT: zookeeper:2181
    KAFKA_ADVERTISED_LISTENERS: PLAINTEXT://kafka:29092,PLAINTEXT_HOST://localhost:9092
    KAFKA_LISTENER_SECURITY_PROTOCOL_MAP: PLAINTEXT:PLAINTEXT,PLAINTEXT_HOST:PLAINTEXT
    KAFKA_INTER_BROKER_LISTENER_NAME: PLAINTEXT
    KAFKA_OFFSETS_TOPIC_REPLICATION_FACTOR: 1

email-service:
  build: ./email-service
  environment:
   SPRING_KAFKA_BOOTSTRAPSERVERS: kafka:29092
  ports:
   - "8081:8081"
  depends_on:
   - kafka

advertised.listeners: Listeners to publish to ZooKeeper for clients to use, if different than the listeners config property. In IaaS environments, this may need to be different from the interface to which the broker binds. If this is not set, the value for listeners will be used. Unlike listeners it is not valid to advertise the 0.0.0.0 meta-address.

Please note that KAFKA_ADVERTISED_HOST_NAME has been deprecated and it's recommended to use KAFKA_ADVERTISED_LISTENERS instead. For more information about KAFKA_ADVERTISED_LISTENERS check here.

like image 175
lloiacono Avatar answered Oct 11 '22 13:10

lloiacono


This is Apache Kafka 2.4.0.

I'm sharing the low-level code-based findings to shed more light when this WARN message could be printed out and why. That's certainly a misconfiguration of a Kafka cluster. Read on and comment if there's something missing. Thanks!


The WARN message is printed out when the DefaultMetadataUpdater (of NetworkClient) is requested to handle a completed metadata response.

[count] partitions have leader brokers without a matching listener, including [partitions]

It is a warning that corresponds to Errors.LISTENER_NOT_FOUND that has the following default exception text:

There is no listener on the leader broker that matches the listener on which metadata request was processed.

That's on the client side.

Digging deeper you can find that this Errors.LISTENER_NOT_FOUND is used on a Kafka broker when MetadataCache is requested to find partition metadata. That's where you can find just before there's this DEBUG message:

Error while fetching metadata for [topicPartition]: listener [listenerName] not found on leader [leaderBrokerId]

Simply turn the DEBUG logging level for kafka.server.MetadataCache logger and you should see it in the controller broker's logs.

In this particular case, this MetadataCache is used by a broker (via KafkaApis) to handle TopicMetadata request where they say:

// In versions 5 and below, we returned LEADER_NOT_AVAILABLE if a matching listener was not found on the leader.

// From version 6 onwards, we return LISTENER_NOT_FOUND to enable diagnosis of configuration errors.

And at that moment, it's clear that the WARN message in question is for a connection on the listenerName.


In my case, when I was debugging the issue, it turned out that I used SSL://:9093 to connect to a Kafka broker while the partition leader was neither available nor configured to listen to the listeners configuration property.

I used kafka-topics to review the partition configuration and then reviewed the state of partitions in ZooKeeper.

get /brokers/topics/ssl/partitions/0/state
{"controller_epoch":1,"leader":0,"version":1,"leader_epoch":0,"isr":[0]}

I had -1 for the leader, but the isr showed a broker that was simply misconfigured. That's why people reported they fixed the issue by restarting their clusters (to get all the brokers up and running) or fixing the broker ID to the one that worked previously.

like image 24
Jacek Laskowski Avatar answered Oct 11 '22 14:10

Jacek Laskowski


[context]

I am trying to run a docker compose with a kafka client using the registry of https://github.com/wurstmeister/kafka-docker I am trying to run a very simple kafka cluster with a single broker and 3 topics with each 1 partition and a replication factor of 1.

this great link explains connectivity for a kafka cluster with one broker, a kafka cluster with several brokers and also notions about the listeners, all using docker, please have a look : https://github.com/wurstmeister/kafka-docker/wiki/Connectivity

[result]

The first time i run docker-compose up --force-recreate --build, everything runs just fine !
The topics are created automatically using KAFKA_CREATE_TOPICS and I can use kafka producer and consumer just fine.

list topics : bin/kafka-topics.sh --list --bootstrap-server localhost:9092
producer : bin/kafka-console-producer.sh --broker-list localhost:9092 --topic productadvisor_sales_dev
consumer : bin/kafka-console-consumer.sh --bootstrap-server localhost:9092 --topic productadvisor_sales_dev --from-beginning

After that, everytime I do docker-compose stop, and relaunch using docker-compose up --force-recreate --build and try to produce data I get the following error message ...

Error Message : [2019-09-23 19:41:33,037] WARN [Producer clientId=console-producer] 1 partitions have leader brokers without a matching listener, including [productadvisor_purchase_dev-0] (org.apache.kafka.clients.NetworkClient)

[Answer]

It appears you need to specify the value of KAFKA_BROKER_ID (=1 for instance) so that the zookeeper doesn't try to create a new broker which can't have a listener because it is binded to the old one.

[Code]


services:
  zookeeper:
    image: wurstmeister/zookeeper
    ports:
      - "2181:2181"

  kafka:
    image: wurstmeister/kafka
    ports:
      - "9092:9092"
    environment:
      KAFKA_BROKER_ID: 1
      KAFKA_HOST: localhost
      KAFKA_PORT: 9092
      KAFKA_ADVERTISED_HOST_NAME: localhost
      KAFKA_ADVERTISED_PORT: 9092
      KAFKA_ZOOKEEPER_CONNECT: zookeeper:2181
      KAFKA_CREATE_TOPICS: "productadvisor_sales_dev:1:1,productadvisor_stock_dev:1:1,productadvisor_purchase_dev:1:1"
    depends_on:
      - zookeeper
    command: [start-kafka.sh]

[Some documentation]

  • https://rmoff.net/2018/08/02/kafka-listeners-explained/
  • https://www.tutorialspoint.com/apache_kafka/apache_kafka_fundamentals.htm
  • http://www.michael-noll.com/blog/2013/03/13/running-a-multi-broker-apache-kafka-cluster-on-a-single-node/
  • https://blog.k2datascience.com/running-kafka-using-docker-332207aec73c

NB

if anyone has more information about the inner working of kafka, the zookeeper and the broker and why we need to specify it, why the information is kept even if I do a --force-recreate --build ... please do not hesitate. I am new to kafka and this is one of my first complete post on stackoverflow :)

Cheers !

like image 13
IanMars Avatar answered Oct 11 '22 15:10

IanMars