[ https://issues.apache.org/jira/browse/FLINK-10245?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16615552#comment-16615552 ]
Hequn Cheng commented on FLINK-10245: ------------------------------------- [~dangdangdang] Yes, you are right. We can't simply deduplicate rows according to the rowkey if nullable field is allowed. In this case, we should merge rows in the buffer. When merge two rows, return the new field if it is not null or return the old field. However, there are some problems for UpsertTableSink which contains key definition. In this case, we should stick to the latest value even if it is null. We may need to delete before writing to make sure this semantic, since hbase doesn't support writing null values. We can have more discussions about TableSink in the TabkeSink task(FLINK-10277). What do you think? > Add DataStream HBase Sink > ------------------------- > > Key: FLINK-10245 > URL: https://issues.apache.org/jira/browse/FLINK-10245 > Project: Flink > Issue Type: Sub-task > Components: Streaming Connectors > Reporter: Shimin Yang > Assignee: Shimin Yang > Priority: Major > Labels: pull-request-available > > Design documentation: > [https://docs.google.com/document/d/1of0cYd73CtKGPt-UL3WVFTTBsVEre-TNRzoAt5u2PdQ/edit?usp=sharing] -- This message was sent by Atlassian JIRA (v7.6.3#76005)