[
https://issues.apache.org/jira/browse/KAFKA-12455?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17301952#comment-17301952
]
Ron Dagostino commented on KAFKA-12455:
---------------------------------------
I looked at the brokers' metadata caches for the two separate configurations --
ZK vs. Raft -- to find out what percentage of the time they showed showed 1
alive broker instead of 2. I was expecting the ZooKeeper configuration to show
relatively little time with just 1 alive broker since the clients are never
seeing that metadata situation, and I was expecting the Raft configuration to
show a much higher percentage of time with just 1 alive broker since the
clients do see that metadata situation. I did not find what I was expecting to
find.
The amounts of times where the brokers are advertising just 1 alive broker in
their metadata cache as follows:
*ZooKeeper Configuration*:
BrokerId=1: 37 seconds out of 61 seconds of that broker's availability
during the test, or 61% of the time with just 1 alive broker in metadata cache
BrokerId=2: 39 seconds out of 61 seconds of that broker's availability
during the test, or 64% of the time with just 1 alive broker in metadata cache
*Raft Configuration*:
BrokerId=1: 37 seconds out of 88 seconds of that broker's availability
during the test, or 42% of the time with just 1 alive broker in metadata cache
BrokerId=2: 52 seconds out of 88 seconds of that broker's availability
during the test, or 59% of the time with just 1 alive broker in metadata cache
So the brokers in the Zookeeper configuration consider just 1 broker to be
alive more often than the brokers in the Raft configuration consider just 1
broker to be alive!
It is still not clear why the consumers never see just a single alive broker in
the ZooKeeper configuration. From the above it does not appear to be due to
any difference in metadata cache population -- if it were just that then we
would see the test failing in the ZooKeeper configuration since that actually
advertises a single alive broker more frequently in terms of percentage of test
time.
> OffsetValidationTest.test_broker_rolling_bounce failing for Raft quorums
> ------------------------------------------------------------------------
>
> Key: KAFKA-12455
> URL: https://issues.apache.org/jira/browse/KAFKA-12455
> Project: Kafka
> Issue Type: Bug
> Affects Versions: 2.8.0
> Reporter: Ron Dagostino
> Assignee: Ron Dagostino
> Priority: Blocker
>
> OffsetValidationTest.test_broker_rolling_bounce in `consumer_test.py` is
> failing because the consumer group is rebalancing unexpectedly.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
