[
https://issues.apache.org/jira/browse/KAFKA-7227?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Genmao Yu updated KAFKA-7227:
-----------------------------
Description:
In current behavior, if partial of broker machines have "server.rack", we will
failed to create topic with exception like "Not all brokers have rack
information for replica rack aware assignment.". There is no big problem in
local cluster. But, this behavior may not be very suitable for cloud
environment. In cloud environment, there may be different types of machine,
some has rack info, while others do not has rack info. It could happen in our
kafka cluster for many reasons, like understock of specific machine.
So, We can consider to give each broker a default rack id.
* if not set any rack info, it is the same with previous behavior
* if partial of brokers are set with rack info, we can think of the remaining
brokers as same rack.
As far as know, if not set rack id, DataNode/NodeManager has a default rack id
in Hadoop.
was:
In current behavior, if partial of broker machines have "server.rack", we will
failed to create topic with exception like "Not all brokers have rack
information". There is no big problem in local cluster. But, this behavior may
not be very suitable for cloud environment. In cloud environment, there may be
different types of machine, some has rack info, while others do not has rack
info. It could happen in our kafka cluster for many reasons, like understock of
specific machine.
So, We can consider to give each broker a default rack id.
* if not set any rack info, it is the same with previous behavior
* if partial of brokers are set with rack info, we can think of the remaining
brokers as same rack.
As far as know, if not set rack id, DataNode/NodeManager has a default rack id
in Hadoop.
> Kafka cluster should has a default rack id if not set
> -----------------------------------------------------
>
> Key: KAFKA-7227
> URL: https://issues.apache.org/jira/browse/KAFKA-7227
> Project: Kafka
> Issue Type: Improvement
> Components: core
> Affects Versions: 1.0.0, 2.0.0
> Reporter: Genmao Yu
> Priority: Major
>
> In current behavior, if partial of broker machines have "server.rack", we
> will failed to create topic with exception like "Not all brokers have rack
> information for replica rack aware assignment.". There is no big problem in
> local cluster. But, this behavior may not be very suitable for cloud
> environment. In cloud environment, there may be different types of machine,
> some has rack info, while others do not has rack info. It could happen in our
> kafka cluster for many reasons, like understock of specific machine.
> So, We can consider to give each broker a default rack id.
> * if not set any rack info, it is the same with previous behavior
> * if partial of brokers are set with rack info, we can think of the
> remaining brokers as same rack.
> As far as know, if not set rack id, DataNode/NodeManager has a default rack
> id in Hadoop.
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
