[
https://issues.apache.org/jira/browse/KAFKA-3729?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Richard Yu updated KAFKA-3729:
------------------------------
Labels: api needs-dicussion needs-kip newbie (was: api needs-kip newbie)
> Auto-configure non-default SerDes passed alongside the topology builder
> ------------------------------------------------------------------------
>
> Key: KAFKA-3729
> URL: https://issues.apache.org/jira/browse/KAFKA-3729
> Project: Kafka
> Issue Type: Improvement
> Components: streams
> Reporter: Fred Patton
> Assignee: Ted Yu
> Priority: Major
> Labels: api, needs-dicussion, needs-kip, newbie
> Attachments: 3729.txt, 3729.v6.txt
>
>
> From Guozhang Wang:
> "Only default serdes provided through configs are auto-configured today. But
> we could auto-configure other serdes passed alongside the topology builder as
> well."
> After the PR was merged, we realized that the current approach to implement
> this features is actually not backward compatible. Thus, we need to revert
> the commit for now to not break backward compatibility in 2.3 release. After
> some more thinking, it seems that this feature is actually more complicated
> to get right as it seem on the surface and hence it would required a proper
> KIP.
> The following issues are identified:
> * in the new code, configure() would be called twice, one in user code (if
> people don't rewrite existing applications) and later via Kafka Streams – the
> second call could "reconfigure" the Serde and overwrite the correct
> configuration from the first call done by the user
> * if there are multiple Serdes using the same configuration parameters
> names, it's only possible to specify this parameter name once in the global
> StreamsConfig; hence, it's not possible for users to configure both Serdes
> differently
> * basically, the global StreamsConfig needs to contain all configuration
> parameters over all used Serdes to make a potential second call to
> `configure()` idempotant
> To address the issues, some ideas would be:
> * pass in the configuration via the constructor and deprecate `configure()`
> method
> * add a new method `isConfigured()` that would allow to skip the second
> configuration call within Kafka Streams runtime
> There might be other ways to address this, and the different options should
> be discussed on the KIP.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
