kaushik srinivas created KAFKA-13176:
----------------------------------------
Summary: frequent ISR shrinks and expansion with default
num.replica.fetchers (1) under very low throughput conditions.
Key: KAFKA-13176
URL: https://issues.apache.org/jira/browse/KAFKA-13176
Project: Kafka
Issue Type: Bug
Reporter: kaushik srinivas
Running a 3 node kafka cluster (2.3.x kafka) with 4 cores of cpu and 4Gi of
memory on a k8s environment.
num.replica.fetchers is configured to 1 (default value).
There are around 15 topics in the cluster and all of them receive a very low
rate of records/sec (less than 100 per second most of the cases).
All the topics have more than 10 partitions and 3 replication each.
min.insync.replicas is set to 2. And producers are run with acks level set to
'all'.
we constantly observer ISR shrinks and expansions for almost each topic
partition continuously. shrinks and expansions are mostly seperated by around 6
to 8 seconds mostly usually.
During these shrinks and expands we see a lot of request time outs at the kafka
producer side for these topics.
any known configuration items we can use to overcome this ?
Confused about the fact of continuous ISR shrinks and expands with very low
throughput topics.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
