'How to Reduce Kafka Active Controller Election Time
There are clusters running in Kubernetes environment on 3 different machines with 3 Kafka brokers and 3 Zookepers.
It may take up to 35 seconds for the new active broker to be elected when the machine where the Active Broker is located is de-energized or power-off.
We measure the election time of the active broker by looking at the time when the metric "cp_kafka_controller_kafkacontroller_activecontrollercount" is 0. Is it normal for Active Broker select time to be this high in Apache Kafka 2.7?
How can we further reduce the active controller election time?
I add below Kafka Zookeeper configuration and Partition numbers
Partition Count: _replicamanager_partitioncount: 1209
Kafka Configs:
configurationOverrides:
offsets.topic.replication.factor: "3"
default.replication.factor: 3
min.insync.replicas: 2
auto.create.topics.enable: false
leader.imbalance.check.interval.seconds: 30
Zookeeper Configs:
tickTime: 2000
syncLimit: 5
initLimit: 10
maxClientCnxns: 60
autoPurgeSnapRetainCount: 3
autoPurgePurgeInterval: 24
Sources
This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.
Source: Stack Overflow
| Solution | Source |
|---|