[ https://issues.apache.org/jira/browse/FLINK-6923?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16058032#comment-16058032 ]
ASF GitHub Bot commented on FLINK-6923: --------------------------------------- Github user zhenzhongxu commented on the issue: https://github.com/apache/flink/pull/4149 Hi @tzulitai. Yes, we do have a use case where we need to disable Flink checkpointing because the time interval checkpointing model does not work with our constraints. We had to trigger Kafka commits by manually taking offset snapshot and commit after sink flushes (parallel source/sink operators are chained together), in this case, the partition offset is not incremented until the thread exits sink operator logic. Now, the only way to make the commit accurate is to expose which partition the in-flight message belongs to and we can consciously +1 to the offset at the time of snapshoting. > Kafka connector needs to expose information about in-flight record in > AbstractFetcher base class > ------------------------------------------------------------------------------------------------ > > Key: FLINK-6923 > URL: https://issues.apache.org/jira/browse/FLINK-6923 > Project: Flink > Issue Type: Improvement > Components: Kafka Connector > Reporter: Zhenzhong Xu > Assignee: Zhenzhong Xu > Priority: Minor > > We have a use case where we have our custom Fetcher implementation that > extends AbstractFetcher base class. We need to periodically get current in > flight (in processing) records' partition and offset information. > This can be easily exposed in AbstractFetcher class. -- This message was sent by Atlassian JIRA (v6.4.14#64029)