[
https://issues.apache.org/jira/browse/KAFKA-13693?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17498090#comment-17498090
]
Augusto Hack commented on KAFKA-13693:
--
[~showuon] here are some additional details:
* Kafka version: 2.8.1
* partitions: 3
* policy: compact
* the znode for the topic was never modified (version was 0, and ctime and
mtime equal)
* brokers existed before the topic was created (checked via timestamp of the
znodes in `/brokers/ids`)
Observed bug:
* The LogConfig object of one of the partitions in the leader had the flag
delete set to true, every other LogConfig had the correct value (true for
compact, false for delete). Checked with a memory dump and querying using
Memory Analyzer (MAT)
> Config in memory got inconsistency value
>
>
> Key: KAFKA-13693
> URL: https://issues.apache.org/jira/browse/KAFKA-13693
> Project: Kafka
> Issue Type: Bug
>Reporter: Aiqin Zhang
>Priority: Minor
> Attachments: Screenshot from 2022-02-25 14-34-36.png
>
>
> We have a kafka cluster with a topic which configured as clean up policy =
> compact.
> But found after timeout timer, the data got deleted on one of the partitions.
>
> Further investigation shows that in memory the value is actually delete which
> is inconsistent.
>
> Topic was created via Terraform in bulk, around 250 together and returned
> with success.
>
> We did the broker restart to correct the setting in memory.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)