[
https://issues.apache.org/jira/browse/KAFKA-8408?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16846157#comment-16846157
]
Jose Armando Garcia Sancio commented on KAFKA-8408:
---------------------------------------------------
What do you think [~cmccabe] [~hachikuji]?
> Create one request purgator for all the controller requests
> -----------------------------------------------------------
>
> Key: KAFKA-8408
> URL: https://issues.apache.org/jira/browse/KAFKA-8408
> Project: Kafka
> Issue Type: Improvement
> Components: controller
> Affects Versions: 2.4
> Reporter: Jose Armando Garcia Sancio
> Assignee: Jose Armando Garcia Sancio
> Priority: Critical
>
> Some RPCs (i.e. leader election) for the controller are going through the
> replica manager. It looks like the reason for this is because currently the
> replica manager has all of the purgatory queues. Another other issue is that
> we have one purgatory per request type; this makes it harder (more work) to
> add requests that go directly to the controller by passing zk.
> In the case of the controller it should be enough to have one purgatory queue
> for all requests. This should make it easier to add more requests for the
> controller. And to migrate all of the requests that go through ZK to instead
> get routed to the controller's event queue.
> To resolve this issue we should:
> 1. create one purgatory can be used for all of the controller's RPC.
> 1. move this purgatory out of the replica manager
> 1. refactor the current leader election purgatory to instead use the generic
> purgatory
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
