[
https://issues.apache.org/jira/browse/FLINK-4789?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16107050#comment-16107050
]
Tzu-Li (Gordon) Tai commented on FLINK-4789:
--------------------------------------------
I would like to close this issue now.
Since Flink 1.4, with partition discovery we will always try to discover new
partitions on restore, and continuous partition discovery will require a
dedicated consumer independent of the one used for record fetching.
> Avoid Kafka partition discovery on restore and share consumer instance for
> discovery and data consumption
> ---------------------------------------------------------------------------------------------------------
>
> Key: FLINK-4789
> URL: https://issues.apache.org/jira/browse/FLINK-4789
> Project: Flink
> Issue Type: Improvement
> Components: Kafka Connector
> Affects Versions: 1.2.0
> Reporter: Robert Metzger
>
> As part of FLINK-4379, the Kafka partition discovery was moved from the
> Constructor to the open() method. This is in general a good change, as
> outlined in FLINK-4155, as it allows us to detect new partitions and topics
> based on regex on the fly.
> However, currently the partitions are discovered on restore as well.
> Also, the {{FlinkKafkaConsumer09.getKafkaPartitions()}} is creating a
> separate {{KafkaConsumer}} just for the partition discovery.
> Since the partition discovery happens on the task managers now, we can use
> the regular {{KafkaConsumer}} instance, which is used for data retrieval as
> well.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
