[
https://issues.apache.org/jira/browse/KAFKA-6316?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Krishan updated KAFKA-6316:
---------------------------
Description:
There is a very weird flow happening in the following case
I have 6 consumer processes running connected to a cluster of 3 brokers. 3 of
the consumers (set A) are subscribed to a particular set of topics (topics C)
and the other 3 consumers (set B) are subscribed to another set of topics
(topics D). All the consumers belong to the same consumer group. All the topics
have a replication factor of 3 and are divided into 24 partitions. Each
consumer is automatically assigned to 8 partitions of the topic set it is
supposed to consume. I have set the assignment pattern to round robin.
If I kill a consumer from set A, topics C are repartitioned successfully in the
remaining consumers from set A. Each consumer in set A is now subscribed to 12
partitions from topics C.
But consumers in set B go crazy. All topics and all partitions in D are
assigned to any one consumer from B and the rest of the consumers remain idle -
All 24 partitions of all topics in set D are assigned to a single process.
I need some idea on how to debug this and what could be the issue?
I believe only the consumers in topic C should be affected. Consumer set B
should remain unaffected and definitely not assign all partitions to a single
process
was:
There is a very weird flow happening in the following case
I have 6 consumer processes running connected to a cluster of 3 brokers. 3 of
the consumers (set A) are subscribed to a particular set of topics (topics C)
and the other 3 consumers (set B) are subscribed to another set of topics
(topics D). All the topics belong to the same consumer group. All the topics
have a replication factor of 3 and are divided into 24 partitions. Each
consumer is automatically assigned to 8 partitions of the topic set it is
supposed to consume. I have set the assignment pattern to round robin.
If I kill a consumer from set A, topics C are repartitioned successfully in the
remaining consumers from set A. Each consumer in set A is now subscribed to 12
partitions from topics C.
But consumers in set B go crazy. All topics and all partitions in D are
assigned to any one consumer from B and the rest of the consumers remain idle -
All 24 partitions of all topics in set D are assigned to a single process.
I need some idea on how to debug this and what could be the issue?
I believe only the consumers in topic C should be affected. Consumer set B
should remain unaffected and definitely not assign all partitions to a single
process
> Kafka Rebalancing issues when I kill one consumer
> -------------------------------------------------
>
> Key: KAFKA-6316
> URL: https://issues.apache.org/jira/browse/KAFKA-6316
> Project: Kafka
> Issue Type: Bug
> Components: consumer
> Affects Versions: 0.11.0.0
> Reporter: Krishan
> Labels: consumer, rebalance
>
> There is a very weird flow happening in the following case
> I have 6 consumer processes running connected to a cluster of 3 brokers. 3 of
> the consumers (set A) are subscribed to a particular set of topics (topics C)
> and the other 3 consumers (set B) are subscribed to another set of topics
> (topics D). All the consumers belong to the same consumer group. All the
> topics have a replication factor of 3 and are divided into 24 partitions.
> Each consumer is automatically assigned to 8 partitions of the topic set it
> is supposed to consume. I have set the assignment pattern to round robin.
> If I kill a consumer from set A, topics C are repartitioned successfully in
> the remaining consumers from set A. Each consumer in set A is now subscribed
> to 12 partitions from topics C.
> But consumers in set B go crazy. All topics and all partitions in D are
> assigned to any one consumer from B and the rest of the consumers remain idle
> - All 24 partitions of all topics in set D are assigned to a single process.
> I need some idea on how to debug this and what could be the issue?
> I believe only the consumers in topic C should be affected. Consumer set B
> should remain unaffected and definitely not assign all partitions to a single
> process
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
