[
https://issues.apache.org/jira/browse/KAFKA-1535?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14059287#comment-14059287
]
Jay Kreps commented on KAFKA-1535:
----------------------------------
Yes I think that is correct. The current behavior is basically an optimization
to avoid sending all the nodes. However this means if you are only sending to a
single partition you may only have the host/port info for one machine, if that
machine goes down you don't have another broker to go to. Since there shouldn't
be more than a few hundred brokers, sending the full host/port for all of them
should be okay and fixes this scenario.
> return all live brokers in TopicMetadataResponse
> ------------------------------------------------
>
> Key: KAFKA-1535
> URL: https://issues.apache.org/jira/browse/KAFKA-1535
> Project: Kafka
> Issue Type: Improvement
> Components: core
> Affects Versions: 0.8.2
> Reporter: Jun Rao
> Labels: newbie
>
> Currently, we only return the brokers that have assigned replicas for a topic
> in TopicMetadataResponse. The new producer will use those brokers for
> refreshing metadata. Now suppose that we stop all those brokers, copy all
> local data to some new hosts and then restart those hosts (with the original
> broker id). There is no way for the new producer to automatically get the
> information about the new brokers since all old brokers are gone.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
