[
https://issues.apache.org/jira/browse/FLINK-23181?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17371450#comment-17371450
]
Aitozi commented on FLINK-23181:
--------------------------------
Thanks [~jark] for looking at this. This issue is meant to add an ability to
extend the current HOP window in SQL.
In many realtime feature job, user have to calculate the recently 1h, 3h, 6h
... accumulators in one sql job. It can be used like
code1:
{code:java}
SELECT * FROM TABLE(
HOP(
DATA => TABLE Bid,
TIMECOL => DESCRIPTOR(bidtime),
SLIDE => INTERVAL '1' MINUTES,
SIZE => INTERVAL '1' HOURS))
{code}
code2:
{code:java}
SELECT * FROM TABLE(
HOP(
DATA => TABLE Bid,
TIMECOL => DESCRIPTOR(bidtime),
SLIDE => INTERVAL '1' MINUTES,
SIZE => INTERVAL '3' HOURS))
{code}
...
But this will lead to redundant storage of window state, In our internal
version, we add support for multi hop in one job , and share the window state
for a better performance. I think it's a common case and I think it can be work
with the current window tvf mode, what's your idea [~jark] ?
> Add support for multi hop window
> ---------------------------------
>
> Key: FLINK-23181
> URL: https://issues.apache.org/jira/browse/FLINK-23181
> Project: Flink
> Issue Type: New Feature
> Components: Table SQL / Planner, Table SQL / Runtime
> Reporter: Aitozi
> Priority: Major
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
