[ https://issues.apache.org/jira/browse/KAFKA-10848?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17268915#comment-17268915 ]
Matthias J. Sax commented on KAFKA-10848: ----------------------------------------- Thanks [~guozhang] – the idea of this ticket was along the same lines as your proposal. How we design the interface would be something to discuss on the KIP. Samza for example uses a quite different interface: [https://samza.apache.org/learn/documentation/0.7.0/api/javadocs/org/apache/samza/system/chooser/MessageChooser.html] – Don't know which API would be favorable. > Allow fine grained control over cross-partition processing order > ---------------------------------------------------------------- > > Key: KAFKA-10848 > URL: https://issues.apache.org/jira/browse/KAFKA-10848 > Project: Kafka > Issue Type: Improvement > Components: streams > Reporter: Matthias J. Sax > Priority: Major > Labels: needs-kip > > Currently, KafkaStreams implements a hard-coded timestamp based strategy to > pick the next record to process for a task, given that a task has multiple > partitions. > In general, this strategy works well for the DSL, but for PAPI users, there > might be cases when the strategy should be customized. And even for the DSL, > there is one corner case (for a stream-table join) for which the table-side > record should be processed first if two records have the same timestamp (at > least, this gap exists as long as we don't have multi-version KTables), while > we cannot enforce this behavior because at runtime we don't know anything > about KStream vs KTable or an existing downstream join. > Thus, we might want to allow users to plugin a custom strategy to pick the > next record for processing. -- This message was sent by Atlassian Jira (v8.3.4#803005)