[ https://issues.apache.org/jira/browse/FLINK-22714?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17358430#comment-17358430 ]
Wenlong Lyu commented on FLINK-22714: ------------------------------------- hi, [~qingru zhang], is this issue depends on other feature not merged? Currently the StreamExecPhysicalWindowTableFunction is an intermediate node in plan optimization(as we can see, StreamExecWindowTableFunction cannot translate to Transformation), I don't see the reason why we need the tag of emitPerRecord on StreamPhysicalWindowTableFunction. > Simplify `StreamPhysicalWindowTableFunction` to a simple window assigner if > successor node with `WindowAttachedWindowingStrategy` in planner > --------------------------------------------------------------------------------------------------------------------------------------------- > > Key: FLINK-22714 > URL: https://issues.apache.org/jira/browse/FLINK-22714 > Project: Flink > Issue Type: Sub-task > Components: Table SQL / Planner > Affects Versions: 1.14.0 > Reporter: JING ZHANG > Assignee: JING ZHANG > Priority: Major > Labels: pull-request-available > > In theory, `StreamPhysicalWindowTableFunction` should works as a window > assigner and emits at ends of window. However, > `StreamPhysicalWindowTableFunction` could be simplified to a simple window > assigner, and emits per record if it's successor node is one of > `WindowRank`/`WindowJoin`/`WindowAggregate` with > `WindowAttachedWindowingStrategy`, because those nodes would assign input > records to windows have been assigned in input nodes, and trigger emit at > ends of window. > Besides, we should add a limit that window is based on rowtime, because it > has syntax problem if window is assigned based on the proc-time of previous > operator. -- This message was sent by Atlassian Jira (v8.3.4#803005)