[
https://issues.apache.org/jira/browse/KAFKA-4633?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16041772#comment-16041772
]
Matthias J. Sax commented on KAFKA-4633:
----------------------------------------
[~agomez] It's by desing. All topic specified by name/patter in the DSL/PAPI
are considered "user topics" and thus, user is responsible to create those
(with appropriate configuration setting) before starting the Streams
application. (cf.
http://docs.confluent.io/current/streams/developer-guide.html#managing-topics-of-a-kafka-streams-application
-- we are still working on AK docs to get this content there, too). If you
don't want to manage intermediate topics you can also rely on
auto-repartitioning that was added in {{0.10.1}}: for this, Streams does create
the topics for you and thus, the topics are no longer considered "user topics"
but "internal topics" (cf KAFKA-3561 and
http://docs.confluent.io/3.1.2/streams/upgrade-guide.html#auto-repartitioning)
> Always use regex pattern subscription to avoid auto create topics
> -----------------------------------------------------------------
>
> Key: KAFKA-4633
> URL: https://issues.apache.org/jira/browse/KAFKA-4633
> Project: Kafka
> Issue Type: Bug
> Components: streams
> Reporter: Guozhang Wang
> Assignee: Guozhang Wang
> Labels: architecture
> Fix For: 0.10.2.0
>
>
> In {{KafkaConsumer}}, a metadata update is requested whenever
> {{subscribe(List<String> topics ..)}} is called. And when such a metadata
> request is sent to the broker upon the first {{poll}} call, it will cause the
> broker to auto-create any topics that do not exist if the broker-side config
> {{topic.auto.create}} is turned on.
> In order to work around this issue until the config is default to false and
> gradually be deprecated, we will let Streams to always use the other
> {{subscribe}} function with regex pattern, which will send the metadata
> request with empty topic list and hence won't trigger broker-side auto topic
> creation.
> The side-effect is that the metadata response will be larger, since it
> contains all the topic infos; but since we only refresh it infrequently this
> will add negligible overhead.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
