[ https://issues.apache.org/jira/browse/FLINK-3581?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15266754#comment-15266754 ]
ASF GitHub Bot commented on FLINK-3581: --------------------------------------- Github user aljoscha commented on the pull request: https://github.com/apache/flink/pull/1957#issuecomment-216261831 I had a version that was doing this. The problem there is that you don't know when you can drop column families. For knowing when a column family is empty you would have to keep a Set of keys for which you have state in memory. Which somewhat defeats the purpose of the RocksDB backend. > Add Special Aligned Event-Time WindowOperator > --------------------------------------------- > > Key: FLINK-3581 > URL: https://issues.apache.org/jira/browse/FLINK-3581 > Project: Flink > Issue Type: Improvement > Components: Streaming > Reporter: Aljoscha Krettek > Assignee: Aljoscha Krettek > > The current Window Trigger is per key. Meaning every window has a (logical) > Trigger for every key in the window, i.e. there will be state and time > triggers per key per window. > For some types of windows, i.e. based on time it is possible to use a single > Trigger to fire for all keys at the same time. In that case we would save a > lot of space on state and timers. Which makes state snapshots a lot smaller. -- This message was sent by Atlassian JIRA (v6.3.4#6332)