[ https://issues.apache.org/jira/browse/FLINK-8951?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16596447#comment-16596447 ]
Sergey Tsvetkov commented on FLINK-8951: ---------------------------------------- I'd like to give this one a try. > Support OVER windows PARTITION BY (rounded) timestamp > ----------------------------------------------------- > > Key: FLINK-8951 > URL: https://issues.apache.org/jira/browse/FLINK-8951 > Project: Flink > Issue Type: New Feature > Components: Table API & SQL > Reporter: Fabian Hueske > Priority: Minor > > There are a few interesting use cases that can be addressed by queries that > follow the following pattern > {code:sql} > SELECT sensorId COUNT(*) OVER (PARTITION BY CEIL(rowtime TO HOUR) ORDER BY > temp ROWS BETWEEN UNBOUNDED preceding AND CURRENT ROW) FROM sensors > {code} > Such queries can be used to compute rolling cascading (tumbling) windows with > aggregates that are reset in regular intervals. This can be useful for TOP-K > per minute/hour/day queries. > Right now, such {{OVER}} windows are not supported, because we require that > the {{ORDER BY}} clause is defined on a timestamp (time indicator) attribute. > In order to support this kind of queries, we would require that the > {{PARTITION BY}} clause contains a timestamp (time indicator) attribute or a > function that is defined on it and which is monotonicity preserving. Once the > optimizer identifies this case, it could translate the query into a special > time-partitioned OVER window operator. -- This message was sent by Atlassian JIRA (v7.6.3#76005)