Ara Zarifian created KAFKA-12150:
------------------------------------
Summary: Consumer group refresh not working with clustered MM2
setup
Key: KAFKA-12150
URL: https://issues.apache.org/jira/browse/KAFKA-12150
Project: Kafka
Issue Type: Bug
Components: mirrormaker
Affects Versions: 2.7.0
Reporter: Ara Zarifian
I'm running MM2 with Kafka 2.7 with the following configuration:
{code}
clusters = eastus2, westus
eastus2.bootstrap.servers = clusrter1.example.com:9092
westus.bootstrap.servers = cluster2.example.com:9092
eastus2->westus.enabled = true
eastus2->westus.topics = .*
westus->eastus2.enabled = true
westus->eastus2.topics = .*
refresh.topics.enabled = true
refresh.topics.interval.seconds = 5
refresh.groups.enabled = true
refresh.groups.interval.seconds = 5
sync.topic.configs.enabled = true
sync.topic.configs.interval.seconds = 5
sync.topic.acls.enabled = false
sync.topic.acls.interval.seconds = 5
sync.group.offsets.enabled = true
sync.group.offsets.interval.seconds = 5
emit.checkpoints.enabled = true
emit.checkpoints.interval.seconds = 5
emit.heartbeats.enabled = true
emit.heartbeats.interval.seconds = 5
replication.factor = 3
checkpoints.topic.replication.factor = 3
heartbeats.topic.replication.factor = 3
offset-syncs.topic.replication.factor = 3
offset.storage.replication.factor = 3
status.storage.replication.factor = 3
config.storage.replication.factor = 3
{code}
More specifically, I'm running multiple instances of MM2 with the above
configuration within Kubernetes pods. I was testing the new automatic consumer
group offset translation functionality and noticed what appears to be a problem
when running more than 1 instance of MM2 in this fashion.
Based on [on the
KEP|https://cwiki.apache.org/confluence/display/KAFKA/KIP-382%3A+MirrorMaker+2.0],
I should be able to run multiple instances in this manner (see "Running a
dedicated MirrorMaker cluster"), however, I noticed that when enabling
replication using a 3-instance MM2 cluster, consumer groups were not
synchronizing across clusters at all.
When running through my test case with a single MM2 instance, consumer group
synchronization appears to work as expected consistently. When running through
my 3-node test case, synchronization begins as soon as I scale the number of
replicas to 1.
Am I misinterpreting the manner in which the KEP describes MM2 clusters or is
this interaction an unexpected one?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
