[
https://issues.apache.org/jira/browse/FLINK-22743?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
fhan updated FLINK-22743:
-------------------------
Description:
assume that we simply do some calculations using TUMBLE* functions like:
"select name,
TUMBLE_START(ts, INTERVAL '24' HOUR),
TUMBLE_END(ts, INTERVAL '24' HOUR),
count(\*)
from source_table group by name, TUMBLE(ts, INTERVAL '24' HOUR);"
when ts = 2021-05-20 12:00:00,
the time of window start is 2021-05-20 08:00:00, and the end is 2021-05-21
08:00:00.
This is not flexible when we want to change the start or end time of the
windows. for example, if we want to calculate from the middle of a day, so we
need the window start time is 2021-05-20 12:00:00 and the end time is
2021-05-21 12:00:00.
we can add offset for TUMBLE* function for meeting these requirements.
for example, the query maybe:
"select name,
TUMBLE_START(ts, INTERVAL '24' HOUR, INTERVAL '4' HOUR),
TUMBLE_END(ts, INTERVAL '24' HOUR, INTERVAL '4' HOUR),
count(\*)
from source_table group by name, TUMBLE(ts, INTERVAL '24' HOUR, INTERVAL '4'
HOUR);"
the third parameter( INTERVAL '4' HOUR) here is the offset for TUMBLE*
functions.
This is a real requirement in our business, so we wonder whether this is a
common requirement and whether our solution make sense.
welcome for discussion.
was:
assume that we simply do some calculations using TUMBLE* functions like:
"select name,
TUMBLE_START(ts, INTERVAL '24' HOUR),
TUMBLE_END(ts, INTERVAL '24' HOUR),
count(*)
from source_table group by name, TUMBLE(ts, INTERVAL '24' HOUR);"
when ts = 2021-05-20 12:00:00,
the time of window start is 2021-05-20 08:00:00, and the end is 2021-05-21
08:00:00.
This is not flexible when we want to change the start or end time of the
windows. for example, if we want to calculate from the middle of a day, so we
need the window start time is 2021-05-20 12:00:00 and the end time is
2021-05-21 12:00:00.
we can add offset for TUMBLE* function for meeting these requirements.
for example, the query maybe:
"select name,
TUMBLE_START(ts, INTERVAL '24' HOUR, INTERVAL '4' HOUR),
TUMBLE_END(ts, INTERVAL '24' HOUR, INTERVAL '4' HOUR),
count(*)
from source_table group by name, TUMBLE(ts, INTERVAL '24' HOUR, INTERVAL '4'
HOUR);"
the third parameter( INTERVAL '4' HOUR) here is the offset for TUMBLE*
functions.
This is a real requirement in our business, so we wonder whether this is a
common requirement and whether our solution make sense.
welcome for discussion.
> Adding offsets for TUMBLE* functions for more flexible window start and end
> time.
> ---------------------------------------------------------------------------------
>
> Key: FLINK-22743
> URL: https://issues.apache.org/jira/browse/FLINK-22743
> Project: Flink
> Issue Type: Improvement
> Components: Table SQL / Planner
> Reporter: fhan
> Priority: Minor
> Labels: flinksql
>
> assume that we simply do some calculations using TUMBLE* functions like:
> "select name,
> TUMBLE_START(ts, INTERVAL '24' HOUR),
> TUMBLE_END(ts, INTERVAL '24' HOUR),
> count(\*)
> from source_table group by name, TUMBLE(ts, INTERVAL '24' HOUR);"
> when ts = 2021-05-20 12:00:00,
> the time of window start is 2021-05-20 08:00:00, and the end is 2021-05-21
> 08:00:00.
> This is not flexible when we want to change the start or end time of the
> windows. for example, if we want to calculate from the middle of a day, so
> we need the window start time is 2021-05-20 12:00:00 and the end time is
> 2021-05-21 12:00:00.
>
> we can add offset for TUMBLE* function for meeting these requirements.
> for example, the query maybe:
> "select name,
> TUMBLE_START(ts, INTERVAL '24' HOUR, INTERVAL '4' HOUR),
> TUMBLE_END(ts, INTERVAL '24' HOUR, INTERVAL '4' HOUR),
> count(\*)
> from source_table group by name, TUMBLE(ts, INTERVAL '24' HOUR, INTERVAL '4'
> HOUR);"
>
> the third parameter( INTERVAL '4' HOUR) here is the offset for TUMBLE*
> functions.
>
> This is a real requirement in our business, so we wonder whether this is a
> common requirement and whether our solution make sense.
> welcome for discussion.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
