[ https://issues.apache.org/jira/browse/FLINK-11944?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Kurt Young updated FLINK-11944: ------------------------------- Component/s: (was: API / Table SQL) SQL / Planner Runtime / Operators > Support FirstRow and LastRow for SQL > ------------------------------------ > > Key: FLINK-11944 > URL: https://issues.apache.org/jira/browse/FLINK-11944 > Project: Flink > Issue Type: New Feature > Components: Runtime / Operators, SQL / Planner > Reporter: Jark Wu > Priority: Major > > Usually there are some duplicate data in the source due to some reasons. In > order to get a correct result, we need to do deduplication. FirstRow and > LastRow are two different strategy for deduplication. The syntax of FirstRow > and LastRow is similar to TopN, but order by a time attribute. For example: > SELECT * > FROM ( > SELECT *, > ROW_NUMBER() OVER (PARTITION BY id ORDER BY proctime DESC) as rownum > FROM T > ) > WHERE rownum = 1 > Some information about FirstRow & LastRow. > 1. the partition by key is the deduplicate key > 2. can only order by a time attribute (either proctime or rowtime) > 3. the rownum filter must be {{= 1}} or {{<= 1}} > 4. it is FirstRow when order direction is ASC, LastRow when order direction > is DESC > This issue is aiming to optimize this query to a FirstLastRow node instead of > Over plus Calc. And translate the it into physical operators. -- This message was sent by Atlassian JIRA (v7.6.3#76005)