[
https://issues.apache.org/jira/browse/KAFKA-18394?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alyssa Huang updated KAFKA-18394:
---------------------------------
Description:
We essentially have two paths of epoch state transitions -
1. one which involves updating on-disk election state and then in-memory epoch
state (which includes in-memory election state changes)
2. one which does not involve any changed election state and just needs an
in-memory epoch state change
For the second case, we should explicitly check no persisted or in-memory
election state has changed. For the first case, we should have the in-memory
election state derive from the persisted election state change. (It should be
impossible for the two to differ)
was:
We essentially have two paths of epoch state transitions -
1. one which involves updating on-disk election state and then in-memory epoch
state (which includes in-memory election state changes)
2. on which does not involve any changed election state and just needs an
in-memory epoch state change
For the second case, we should explicitly check no persisted or in-memory
election state has changed. For the first case, we should have the in-memory
election state derive from the persisted election state change. (It should be
impossible for the two to differ)
> Formalize durable and in-memory election state changes
> ------------------------------------------------------
>
> Key: KAFKA-18394
> URL: https://issues.apache.org/jira/browse/KAFKA-18394
> Project: Kafka
> Issue Type: Improvement
> Reporter: Alyssa Huang
> Assignee: Jose Armando Garcia Sancio
> Priority: Major
>
> We essentially have two paths of epoch state transitions -
> 1. one which involves updating on-disk election state and then in-memory
> epoch state (which includes in-memory election state changes)
> 2. one which does not involve any changed election state and just needs an
> in-memory epoch state change
>
> For the second case, we should explicitly check no persisted or in-memory
> election state has changed. For the first case, we should have the in-memory
> election state derive from the persisted election state change. (It should be
> impossible for the two to differ)
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
