Topic can't be found when producing messages: UNKNOWN_TOPIC_OR_PARTITION

Question

I have a two-nodes Kafka cluster (EC2 instances) where each node is used as a separate broker. When I run a producer on the leader instance with the following command:

kafka-console-producer.sh   --broker-list localhost:9092 --topic test

I get the following errors.

test message [2017-01-09 13:22:39,483] WARN Error while fetching metadata with correlation id 0 : {test=UNKNOWN_TOPIC_OR_PARTITION} (org.apache.kafka.clients.NetworkClient) [2017-01-09 13:22:39,562] WARN Error while fetching metadata with correlation id 1 : {test=UNKNOWN_TOPIC_OR_PARTITION} (org.apache.kafka.clients.NetworkClient) [2017-01-09 13:22:39,663] WARN Error while fetching metadata with correlation id 2 : {test=UNKNOWN_TOPIC_OR_PARTITION} (org.apache.kafka.clients.NetworkClient) ...

Listing the topics with kafka-topics.sh shows that the topic exists.

The description of the topic with:

kafka-topics.sh --zookeeper localhost:2181 --describe --topic test

returns

 Topic:test PartitionCount:8    ReplicationFactor:1 Configs:
     Topic: test    Partition: 0    Leader: 1   Replicas: 1 Isr: 1
     Topic: test    Partition: 1    Leader: 2   Replicas: 2 Isr: 2
     Topic: test    Partition: 2    Leader: 1   Replicas: 1 Isr: 1
     Topic: test    Partition: 3    Leader: 2   Replicas: 2 Isr: 2
     Topic: test    Partition: 4    Leader: 1   Replicas: 1 Isr: 1
     Topic: test    Partition: 5    Leader: 2   Replicas: 2 Isr: 2
     Topic: test    Partition: 6    Leader: 1   Replicas: 1 Isr: 1
     Topic: test    Partition: 7    Leader: 2   Replicas: 2 Isr: 2

I am using Kafka 0.10.1.1.

server.properties file contains:

listeners=PLAINTEXT://0.0.0.0:9092
advertised.listeners=PLAINTEXT://0.0.0.0:9092
port=9092
host.name=kafka-node1(kafka-node1 for the second host)
advertised.host.name=kafka-node1(kafka-node2 for the second host)
advertised.port=9092

When I try to produce messages from the second host I get the message below:

WARN Got error produce response with correlation id 1 on topic-partition test-4, retrying (2 attempts left). Error: NOT_LEADER_FOR_PARTITION (org.apache.kafka.clients.producer.internals.Sender) ....

Can anyone please help?

Try to specify all brokers in `--broker-list`. This error may appear when producer tries to send a message to a partition that is not served by a broker. AFAIK, `advertised.*` settings are deprecated. — Yuriy Tseretyan, Jan 09 '17 at 16:14
When I put both hosts in `--broker-list` I still get the same error. — saloua, Jan 10 '17 at 08:24
You don't need port, host.name, advertised.host.name and advertised.port settings, just use listers and advertised.listeners only if clients outside the machine see an address different than the address where service is listening on. Then, for your setup try using broker-list equal to private IP of the machine where broker is running. — Luciano Afranllie, Jan 10 '17 at 19:22
This error message also happen if sentry is active and the user has not the correct permissions. — ozw1z5rd, Oct 23 '18 at 13:47
how about `kafka-console-producer.sh --broker-list kafka-node1:9092 --topic test` ? — Svend, Nov 25 '19 at 19:34
Also, it seems the broker on the node 2 might have difficulties reaching the broker on node 1. I see that because it seems all your partitions with a leader on node 2 have 2 in-sync replicas, which is good, and since replication is a fetch mechanism, I gather that means node 1 can fetch data from node 2. However, partitions with leader on node 1 have only 1 in-sync replica => I guess node 2 fails to fetch from node 1. Try to see by which hostname broker know each other and if the name resolution and routing all goes well — Svend, Nov 25 '19 at 19:41
Though are you sure you want to spend your precious time hosting a kafka yourself? There are tons of hosted Kafka offering on AWS nowadays. AWS itself has MSK (kakfa 2.2.1), Confluent offers Confluent cloud and Aiven is also a very decent choice, among others. Save yourself the trouble, use some hosted kafka :) — Svend, Nov 25 '19 at 19:44

score 1 · Answer 1 · answered May 01 '21 at 18:32

If you are getting UNKNOWN_TOPIC_OR_PARTITION then add the below properties to server.properties file :

listeners=PLAINTEXT://host.name:port
advertised.listeners=PLAINTEXT://host.name:port

in my cases values were as follows:

listeners=PLAINTEXT://localhost:9092
advertised.listeners=PLAINTEXT://localhost:9092

score 0 · Answer 2 · answered Jun 04 '21 at 19:19

Your topic is marked for deletion:

KAFKA_OPTS="-Djava.security.auth.login.config=/home/kafka/kafka_2.11-1.0.0/config/jaas.conf -Djava.security.krb5.conf=/etc/krb5.conf" \
    bin/kafka-topics.sh --describe \
    --zookeeper zookeeper-server-01.eigenroute.com:2181,zookeeper-server-02.eigenroute.com:2181,zookeeper-server-03.eigenroute.com:2181/apps/kafka-cluster-demo
Topic:another-test-topic    PartitionCount:9    ReplicationFactor:2 Configs:    MarkedForDeletion:true
    Topic: another-test-topic   Partition: 0    Leader: none    Replicas: 2,1   Isr:
    Topic: another-test-topic   Partition: 1    Leader: none    Replicas: 1,2   Isr:
    Topic: another-test-topic   Partition: 2    Leader: none    Replicas: 2,1   Isr:
...

score -4 · Answer 3 · edited Jan 08 '18 at 01:38

-4

Make the replication factor 2 for topic test and you won't get this exception anymore. Since you have a 2 broker cluster, execute the command below to produce messages:

kafka-console-producer.sh   --broker-list localhost:9092,localhost:9093 --topic test

This will send messages to both of the brokers in the clusters.

edited Jan 08 '18 at 01:38

Shafique Jamal

1,550
3
21
45

answered Apr 18 '17 at 08:07

shiba goyal

9
3

4

Kafka producers send every message to servers which has the partition leader for that particular topic. So your last sentence is incorrect. broker list is used for initial discovery. – YamYamm Feb 20 '18 at 14:03
1

Does this sollution solvers the problem? To me this does not seem correct. If we had a cluster with 50 nodes we would need to make replication factor equal to 50? – giannisapi Jun 08 '18 at 12:23
I agree with @giannisapi above - what is the correct solution for this? – Cory Robinson Sep 05 '19 at 19:47
2

@CoryRobinson I can not know what is the problem in your case but the problem seems to be in the communication between the brokers or between the brokers and the zookeepers. Check the logs of all the zookeeper and kafka nodes and it will give you a hint. – giannisapi Sep 06 '19 at 06:33

Topic can't be found when producing messages: UNKNOWN_TOPIC_OR_PARTITION

3 Answers3

Linked