[
https://issues.apache.org/jira/browse/KAFKA-19347?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Lianet Magrans resolved KAFKA-19347.
------------------------------------
Resolution: Fixed
Closing as I see the fix and follow-up merged
> AclControlManager unsafe state update
> -------------------------------------
>
> Key: KAFKA-19347
> URL: https://issues.apache.org/jira/browse/KAFKA-19347
> Project: Kafka
> Issue Type: Bug
> Components: controller
> Reporter: David Arthur
> Assignee: David Arthur
> Priority: Major
>
> In AclControlManager#createAcls, we are updating the timeline data structures
> prior to the record being replayed. This could potentially allow uncommitted
> state to be exposed to callers. In general, this kind of update could lead to
> data consistency problems, but for ACLs in particular there is actually no
> problem.
>
> On the write side of things, createAcls was making the same update as the
> replay method and this action is idempotent. This means the end result will
> be the same.
>
> On the read side, the concern is that uncommitted state is acted on and the
> controller crashes. For ACLs, this actually doesn't present a problem because
> we only apply ACLs from AclsDelta which is populated by MetadataLoader. Since
> MetadataLoader only replays committed records, there is no potential for
> exposing uncommitted data.
> Regardless, we should fix this unsafe update in the controller.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
